Skip to main content

Full text of "Ultrasonic tissue characterization II"

See other formats 


A111D3  D7DbDb 


NAFL  INST  OF  STANDARDS  &  TECH  R.I.C. 


A1 11 03070606 

Sympos/Ultrasonic  tissue  c 
QC100  .U57  N0.525,  1979  C.2  NBS-PUB-C  19 
CO 
Q 


NBS  SPECIAL  PUBLICATION 


525 


U.S.  DEPARTMENT  OF  COMMERCE/National  Bureau  of  Standards 


National  Bureau  of  Standards 
Library,   E-01  Admin.  B!dg. 


OCT  1  1981 


ISICGS 


Ultrasonic 
Tissue  Characterization  II 


NATIONAL  BUREAU  OF  STANDARDS 


The  National  Bureau  of  Standards'  was  established  by  an  act  of  Congress  March  3,  1901.  The 
Bureau's  overall  goal  is  to  strengthen  and  advance  the  Nation's  science  and  technology  and 
facilitate  their  effective  application  for  public  benefit.  To  this  end,  the  Bureau  conducts 
research  and  provides:  (1)  a  basis  for  the  Nation's  physical  measurement  system,  (2)  scientific 
and  technological  services  for  industry  and  government,  (3)  a  technical  basis  for  equity  in 
trade,  and  (4)  technical  services  to  promote  public  safety.  The  Bureau's  technical  work  is 
performed  by  the  National  Measurement  Laboratory,  the  National  Engineering  Laboratory, 
and  the  Institute  for  Computer  Sciences  and  Technology. 

THE  NATIONAL  MEASURE\iENT  LABORATORY  provides  the  national  system  of 
physical  and  chemical  and  materials  measurement;  cooidinates  the  system  with  measurement 
systems  of  other  nations  and  furnishes  essential  services  leading  to  accurate  and  uniform 
physical  and  chemical  measurement  throughout  the  Nation's  scientific  community,  industry, 
and  commerce;  conducts  materials  research  leading  to  improved  methods  of  measurement, 
standards,  and  data  on  the  properties  of  materials  needed  by  industry,  commerce,  educational 
institutions,  and  Government;  provides  advisory  and  research  services  to  other  Government 
Agencies;  develops,  produces,  and  distributes  Standard  Reference  Materials;  and  provides 
calibration  services.  The  Laboratory  consists  of  the  following  centers: 

Absolute  Physical  Quantities^  —  Radiation  Research  —  Thermodynamics  and 
Molecular  Science  —  Analytical  Chemistry  —  Materials  Science. 

THE  NATIONAL  ENGINEERING  LABORATORY  provides  technology  and  technical 
services  to  users  in  the  public  and  private  sectors  to  address  national  needs  and  to  solve 
national  problems  in  the  public  interest;  conducts  research  in  engineering  and  applied  science 
in  support  of  objectives  in  these  efforts;  builds  and  maintains  competence  in  the  necessary 
disciplines  required  to  carry  out  this  research  and  technical  service;  develops  engineering  data 
and  measurement  capabilities;  provides  engineering  measurement  traceability  services; 
develops  test  methods  and  proposes  engineering  standards  and  code  changes;  develops  and 
proposes  new  engineering  practices;  and  develops  and  improves  mechanisms  to  transfer 
results  of  its  research  to  the  utlimate  user.  The  Laboratory  consists  of  the  following  centers: 

Applied  Mathematics  —  Electronics  and  Electrical  Engineering^  —  Mechanical 
Engineering  and  Process  Technology'  —  Building  Technology  —  Fire  Research  — 
Consumer  Product  Technology  —  Field  Methods. 

THE  INSTITUTE  FOR  COMPUTER  SCIENCES  AND  TECHNOLOGY  conducts 
research  and  provides  scientific  and  technical  services  to  aid  Federal  Agencies  in  the  selection, 
acquisition,  application,  and  use  of  computer  technology  to  improve  effectiveness  and 
economy  in  Government  operations  in  accordance  with  Public  Law  89-306  (40  U.S.C.  759), 
relevant  Executive  Orders,  and  other  directives;  carries  out  this  mission  by  managing  the 
Federal  Information  Processing  Standards  Program,  developing  Federal  ADP  standards 
guidelines,  and  managing  Federal  participation  in  ADP  voluntary  standardization  activities; 
provides  scientific  and  technological  advisory  services  and  assistance  to  Federal  Agencies;  and 
provides  the  technical  foundation  for  computer-related  policies  of  the  Federal  Government. 
The  Institute  consists  of  the  following  divisions: 

Systems  and  Software  —  Computer  Systems  Engineering  —  Information  Technology. 

'Headquarters  and  Laboratories  at  Gaithersburg,  Maryland,  unless  otherwise  noted; 
mailing  address  Washington, D.C.  20234. 

^Some  divisions  within  the  center  are  located  at  Boulder,  Colorado,  80303. 


The  National  Bureau  of  Standards  was  ceorganized,  effective  Aprii  9,  1978. 


Ultrasonic  Tissue  Characterization  li 

1  -  . 

A  collection  of  reviewed  papers  based  on  talks  presented  at  the  *Q  O  OC) 

Second  International  Symposium  on  Ultrasonic  Tissue  Characterization  -  ^ 

held  at  the  National  Bureau  of  Standards,  Gaithersburg,  Maryland  U  -  1 

June  13-15,  1977  ^  ^ 

Edited  by  ,  . 

Melvin  Linzer 


National  Measurement  Laboratory 
National  Bureau  of  Standards 
Washington,  DC  20234 


Cosponsors  of  Symposium  on  Ultrasonic  Tissue  Characterization: 

National  Bureau  of  Standards 
(National  Measurement  Laboratory) 

National  Science  Foundation 

(Research  Applied  to  National  Needs,  RANN) 

National  Institutes  of  Health 

(Diagnostic  Radiology  Department,  Clinical  Center) 


U.S.  DEPARTMENT  OF  COMMERCE,  Juanita  M.  Kreos,  Secretary 

Jordan  J.  Baruch,  Assistant  Secretary  for  Science  and  Teclinology 

NATIONAL  BUREAU  OF  STANDARDS,  Ernest  Ambler,  Director 


Issued  April  1 979 


Library  of  Congress  Catalog  Card  Number:  79-600026 


National  Bureau  of  Standards  Special  Publication  525 

Nat.  Bur,  Stand.  (U.S.),  Spec.  Publ.  525,  339  pages  (Apr.  1979) 
CODEN:  XNBSAV 


U.S.  GOVERNMENT  PRINTING  OFFICE 
WASHINGTON:  1979 


For  sale  by  the  Superintendent  of  Documents,  U.S.  Government  Printing  Office,  Washington,  D.C.  20402 
Stock  No.  003-003-02058-3  Price  $5.50 
(Add  25  percent  additional  for  other  than  U.S.  mailing) 


FOREWORD 


The  Second  International  Symposium  on  Ultrasonic  Tissue  Characterization 
set  the  stage  for  establishing  this  series  of  meetings  as  the  world's  leading 
forums  for  the  dissemination  of  the  most  recent  and  advanced  research  in  the 
field.    Undoubtedly,  the  Symposium  contributed  significantly  to  the  contemporary 
improvement  in  medical  diagnosis.    This  improvement  has  resulted  from  the  appli- 
cation to  instrumentation  design  of  the  results  of  measurements  of  the  inter- 
actions of  ultrasound  with  tissue  and  from  the  deeper  understanding  of  the 
physical  principles  underlying  these  interactions. 

The  interdicipl inary  approach  of  engineers,  physical  scientists,  physicians, 
and  mathematicians  made  the  Symposium  a  unique  and  fertile  occasion.  This 
collection  of  reviewed  papers  which  describes  the  research  results  presented  at 
the  meeting  makes  it  possible  for  the  benefits  of  this  work  to  be  shared  by  the 
whole  community  of  those  dedicated  to  the  advancement  of  biomedical  ultrasonics 
and  its  application  to  health  care. 

The  National  Bureau  of  Standards  is  pleased  to  be  responsible  for  this 
publication  and  to  have  joined  the  National  Science  Foundation  and  the  National 
Institutes  of  Health  as  cosponsors  of  the  Symposium. 


John  D.  Hoffman 
Director 

National  Measurement  Laboratory 
National  Bureau  of  Standards 


V 


PREFACE 


This  volume  forms  a  comparison  with  Ultrasonic  Tissue  Characterization,  NBS  Special 
Publication  453  (1976).    It  contains  extended  versions  of  43  of  the  54  papers  presented  at 
the  Second  International  Symposium  on  Ultrasonic  Tissue  Characterization  which  was  held  at 
the  National  Bureau  of  Standards  on  June  13-15,  1977.    In  a  departure  from  the  normal 
practice  for  conference  proceedings,  these  papers  were  critically  reviewed  by  experts  in 
the  field. 

In  the  pages  preceding  the  scientific  papers,  an  overview  of  the. meeting  is  presented. 
The  first  article  in  this  chapter  is  a  resume  of  the  proceedings  of  the  entire  Symposium  and 
it  serves  to  put  the  presentations  and  discussions  into  perspective.    This  is  folowed  by  a 
report  summarizing  the  Panel  Discussion  on  Breast  Cancer,  which  was  one  of  the  highlights  of 
the  meeting,  and  finally,  by  the  introductory  talk  given  by  Dr.  A.  J.  Eggers,  Jr.,  who  at 
the  time  of  the  meeting  was  Assistant  Director  for  Research  Application,  National  Science 
Foundation.    The  scientific  papers  presented  at  the  meeting  are  grouped  into  chapters 
devoted  to:    attenuation  and  velocity  I:    mechanisms;  attenuation  and  velocity  II:  methodo- 
logy and  measurements;  scattering  and  attenuation;  scattering;  tumor  Doppler  signatures; 
propagation  through  bone  and  skull;  image  reconstruction;  signal  processing  and  pattern 
recognition;  and  tissue  viability  and  tissue  phantoms.    A  survey  of  velocity  and  attenuation 
data  in  mammalian  tissue  is  included  in  an  appendix. 

The  success  of  the  Symposium  was  due  to  the  dedicated  efforts  of  many  individuals. 
Acknowledgment    is  given  to  the  members  of  the  Program  committee  (F.  Dunn,  A.  C.  Kak, 
J.  F.  Greenleaf,  J.  G.  Miller,  R.  C.  Waag  and  P.  N.  T.  Wells)  and  to  the  other  individuals 
who  served  as  reviewers  for  this  volume.    Special  appreciation  is  due  to  Peter  Wells  for 
his  contributions  as  cochai rperson  of  the  Panel  Discussion  on  Breast  Cancer  and  for  his 
valuable  collaboration  in  preparing  the  summaries  of  the  Symposium  and  of  the  Panel 
Discussion.    The  efforts  of  Ronald  B.  Johnson  and  Sara  Torrence  and  their  conference  staffs 
in  the  organization  and  management  of  the  Symposium  are  gratefully  acknowledged.  Special 
thanks  is  given  to  Rosemary  S.  Maddock  who  has  provided  the  coordination  and  the  editorial 
assistance  in  the  many  phases  of  the  preparation  of  this  volume. 

Melvin  Linzer 

Editor  and 

Symposium  Chairperson 


vi 


ABSTRACT 


The  Second  International  Symposium  on  Ultrasonic  Imaging  and  Tissue  Charac- 
terization was  held  at  the  National  Bureau  of  Standards  on  June  13-15,  1977.  The 
meeting  was  cosponsored  by  the  National  Bureau  of  Standards,  the  National  Science 
Foundation,  and  the  National  Institutes  of  Health.    This  volume  contains  extended 
and  reviewed  papers  based  on  43  of  the  54  talks  presented  at  the  Symposium.  Topics 
covered  include  techniques  for  measurement  of  ultrasonic  tissue  parameters,  the 
dependence  of  tissue  properties  on  physical  and  biological  variables  (e.g.,  ultra- 
sonic frequency,  temperature),  mechanisms  of  ultrasonic  tissue  interactions,  propa- 
gation through  bone  and  skull,  tumor  Doppler  signatures,  computerized  tomography, 
signal  processing  and  pattern  recognition,  and  tissue  phantoms.    A  survey  of  velocity 
and  attenuation  data  in  mammalian  tissue  is  included  in  an  appendix. 

Key  words:    Absorption;  attenuation;  computerized  tomography;  Doppler; 

impedance;  medical  diagnosis;  microscopy;  pattern  recognition; 
scattering;  signal  processing;  tissue  characterization;  tissue 
parameters;  ultrasound;  velocity. 


In  order  to  describe  experiments  adequately,  it  has  been  necessary  to  identify  commercial 

materials  and  equipment  in  this  book.    In  no  case  does  such  identification  imply  recommendation 

or  endorsement  by  the  National  Bureau  of  Standards,  nor  does  it  imply  that  the  material  or 
equipment  is  necessarily  the  best  available  for  the  purpose. 

vi  i 


/ 

J 


CONTENTS 


Page 

Chapter  1.  OVERVIEW 


Report  on  the  Symposium   3 

Report  on  Panel  Discussion:    Ultrasonic  Diagnosis  of  Breast  Cancer    11 

Introductory  Address    15 

Chapter  2.    ATTENUATION  AND  VELOCITY  I:  MECHANISMS 

Elements  of  Tissue  Characterization.    Part  I.    Ultrasonic  Propagation  J 
Properties   19 

Johnston,  R.  L.,  Goss,  S.  A.,  Maynard,  V.,  Brady,  J.  K. , 

Frizzell,  L.  A.,  O'Brien,  W.  D.,  Jr.,  and  Dunn,  F. 

Absorption  of  Sound  in  Tissues  ^  .  .  29 

Carstensen,  E.  L. 

Mechanisms  of  Ultrasonic  Attenuation  in  Soft  Tissue    37 

O'Donnell ,  M.  and  Miller,  J.  G. 

Chapter  3.    ATTENUATION  AND  VELOCITY  II:    METHODOLOGY  AND  MEASUREMENTS 

Elements  of  Tissue  Characterization.    Part  II.    Ultrasonic  Propagation 

Parameter  Measurements    43 

Goss,  S.  A.,  Johnston,  R.  L.,  Maynard,  V.,  Nider,  L., 

Frizzell,  L.  A.,  O'Brien,  W.  D.,  Jr.,  and  Dunn,  F. 

A  Device  for  Measuring  Ultrasonic  Propagation  Velocity  in  Tissue    .  .  .  . .  53 
Sollish,  B.  D. 

Measurement  of  the  Temperature  Dependence  of  the  Velocity  of  Ultra-  \y 

sound  in  Soft  Tissues   57 

Bowen ,  T.,  Connor,  W.  G.,  Nasoni ,  R.  L.,  Pifer,  A.  E.,  and 

Sholes,  R.  R. 

Ultrasonic  Attenuation  in  Normal  and  Ischemic  Myocardium    63 

O'Donnell,  M.,  Mimbs,  J.  W.,  Sobel,  B.  E.,  and  Miller,  J.  G. 

Acoustic  Microscopic  Analysis  of  Myocardium    73 

Yuhas,  D.  E.  and  Kessler,  L.  W. 

Acoustic  Properties  of  Normal  and  Abnormal  Human  Brain    81 

Kremkau,  F.  W.,  McGraw,  C.  P.,  and  Barnes,  R.  W. 

Frequency  Dependent  Attenuation  of  Malignant  Breast  Tumors  Studied 

by  the  Fast  Fourier  Transform  Technique   85 

Fry,  E.  K.,  Sanghvi ,  N.  T.,  Fry,  F.  J.,  and  Gallager,  H.  S. 

Correlation  of  Ultrasonic  Attenuation  with  Connective  Tissue  Content 

in  Breast  Cancers   93 

Kobayashi ,  T. 

The  Attenuation  of  Selected  Soft  Tissue  as  a  Function  of  Frequency    101 

Le  Croissette,  D.  H.,  Heyser,  R.  C,  Gammell,  P.  M., 
Roseboro,  J.  A.,  and  Wilson,  R.  L. 

Chapter  4.    SCATTERING  AND  ATTENUATION 

Clinical  Spectrum  Analysis  Techniques  for  Tissue  Characterization    Ill 

Lizzi,  F.  L.  and  Elbaum,  M.  E. 


ix 


Tissue  Characterization  In  Vivo  by  Differential  Attenuation 

Measurements    121 

L^vi,  S.  and  Keuwez,  J. 

Statistical  Estimation  of  the  Acoustic  Attenuation  Coefficient 

Slope  for  Liver  Tissue  from  Reflected  Ultrasonic  Signals    125 

Kuc,  R.,  Schwartz,  M.,  Finby,  N.,  and  Dain,  F. 

Chapter  5.  SCATTERING 

An  Ultrasonic  Tissue  Signature  for  the  Lung  Surface    135 

Rhyne,  T.  L. 

Angle  Scan  and  Frequency-Swept  Ultrasonic  Scattering  Characteriza- 
tion of  Tissue   143 

Waag,  R.  C,  Lee,  P.  P.  K.,  Lerner,  R.  M.,  Hunter,  L.  P., 

Gramiak,  R.,  and  Schenk,  E.  A. 

Quantitative  Measurements  of  Scattering  of  Ultrasound  by  Heart 

and  Liver   153 

Reid,  J.  M.  and  Shung,  K.  K. 

Dependence  of  Ultrasound  Backscatter  from  Human  Liver  Tissue  on 

Frequency  and  Protein/Lipid  Composition    157 

Freese,  M.  and  Lyons,  E.  A. 

Ultrasound  Backscattering  from  Blood:    Hematocrit  and  Erythrocyte 

Aggregation  Dependence    165 

Hanss,  M.  and  Boynard,  M. 

Chapter  6.    TUMOR  DOPPLER  SIGNATURES 

Tumour  Detection  by  Ultrasonic  Doppler  Blood-Flow  Signals    173 

Wells,  P.  N.  T.,  Halliwell,  M.,  Mountford,  R.  A., 
Skidmore,  R.,  Webb,  A.  J'.,  and  Woodcock,  J.  P. 

Chapter  7.    PROPAGATION  THROUGH  BONE  AND  SKULL 

A  Theory  Relating  Sonic  Velocity  to  Mineral  Content  in  Bone    179 

Lees,  S.  and  Davidson,  C.  L. 

Ultrasonic  Properties  and  Microtexture  of  Human  Cortical  Bone    189 

Yoon,  H.  S. ,  and  Katz,  J.  L. 

Attenuation  and  Dispersion  of  Ultrasound  in  Cancellous  Bone    197 

Barger,  J.  E. 

Transkull  Transmission  of  Axisymmetric  Focused  Ultrasonic  Beams 

in  the  0.5  to  1  MHz  Frequency  Range:    Implications  for  Brain  Tissue 

Visualization,  Interrogation,  and  Therapy    203 

Fry,  F.  J. 

Some  Advances  in  Acoustic  Imaging  Through  Skull    209 

Smith,  S.  W.,  Phillips,  D.  J.,  von  Ramm,  0.  T. ,  and  Thurstone,  F.  L. 

Chapter  8.     IMAGE  RECONSTRUCTION 

Characterization  of  In  Vivo  Breast  Tissue  by  Ultrasonic 

Time-of-Fl ight  Computed  Tomography    221 

Glover,  G.  H. 

Variation  of  Acoustic  Speed  with  Temperature  in  Various  Excised  Human 

Tissues  Studied  by  Ultrasound  Computerized  Tomography    227 

Rajagopalan,  B.,  Greenleaf,  J.  F.,  Thomas,  P.  J.,  Johnson,  S.  A., 

and  Bahn,  R.  C. 


X 


High  Spatial  Resolution  Ultrasonic  Measurement  Techniques  for 

Characterization  of  Static  and  Moving  Tissues    235 

Johnson,  S.  A.,  Greenleaf,  J.  F.,  Rajagopalan,  B.,  Bahn,  R.  C, 
Baxter,  B.  and  Christensen,  D. 

Mapping  True  Ultrasonic  Backscatter  and  Attenuation  Distribution 

in  Tissue  -  A  Digital  Reconstruction  Approach    247 

Duck,  F.  A.  and  Hill,  C.  R. 

Chapter  9.    SIGNAL  PROCESSING  AND  PATTERN  RECOGNITION 

A  Comprehensive  Ultrasonic  Tissue  Analysis  System    255 

Linzer,  M.,  Parks,  S.  I.,  Norton,  S.  J.,  Higgins,  F.  P., 
Dietz,  D.  R.,  Shideler,  R.  W.,  Shawker,  T.  H.,  and 
Doppman,  J.  L. 

Theoretical  Analysis  of  Instantaneous  Power  Spectra  as  Applied 

to  Spectra-Color  Ultrasonography    261 

Jennings,  W.  D.,  Holasek,  E.,  and  Purnell,  E.  W. 

Identification  of  Tissue  Parameters  by  Digital  Processing  of 

Real-Time  Ultrasonic  Clinical  Cardiac  Data    267 

Joynt,  L.,  Boyle,  D.,  Rakowski,  H.,  Popp,  R.,  and  Beaver,  W. 

Dynamic  Autocorrelation  Analysis  of  A-Scans  Iji  Vivo    275 

Gore,  J.  C,  Leeman,  S.,  Metreweli,  C,  Plessner,  N.  J., 
and  Willson,  K.  ■ 

Computer  Spectral  Analysis  of  Ultrasonic  A  Mode  Echoes    281 

Robinson,  D.  E. 

Cepstral  Signal  Processing  for  Tissue  Signature  Analysis    287 

Fraser,  J.,  Kino,  G.  S.,  and  Birnholz,  J. 

Recognition  of  Patterns  in  Ultrasonic  Sectional  Pictures  of  the 

Prostate  for  Tumor  Diagnosis    297 

von  Seelen,  W.,  Gaca,  A.,  Lock,  E.,  Scheiding,  W.,  and 
Wessels,  G. 

Recent  Developments  in  Obtaining  Histopathological  Information  from 

Ultrasound  Tissue  Signatures    303 

Preston,  K.,  Jr.,  Czerwinski,  M.  J.,  Skolnik,  M.  L.,  and 

Leb,  D.  E. 

Chapter  10.    TISSUE  VIABILITY  AND  TISSUE  PHANTOMS 

Damage  and  Death  in  Tissues  and  Associated  Changes  in  Their 

Mechanical  Properties    317 

Weiss,  L. 

A  Human  Abdominal  Tissue  Phantom    323 

Edmonds,  P.  D.,  Reyes,  Z.,  Parkinson,  D.  B.,  Filly,  R.  A., 
and  Busey,  H. 

Tissue  Simulators  for  Diagnostic  Ultrasound    327 

Eggleton,  R.  C.  and  Whitcomb,  J.  A. 

Tissue  Equivalent  Test  Objects  for  Comparison  of  Ultrasound 
Transmission  Tomography  by  Reconstruction  with  Pulse  Echo 

Ultrasound  Imaging    337 

Carson,  P.  L.,  Shabason,  L.,  Dick,  D.  E.,  and  dayman,  W. 

Appendi  X 

Data  of  the  Velocity  and  Attenuation  of  Ultrasound  in 

Mammal ian  Ti ssues  -  A  Survey    343 

Parry,  R.  J.  and  Chivers,  R.  C. 


xi 


I 


Chapter  1 
OVERVIEW 


1 


REPORT  ON  THE  SYMPOSIUM 


The  Second  International  Symposium  on  Ultra- 
',  onic  Tissue  Characterization  had  six  principal 
objectives : 

•to  review  the  progress  which  had  been  made 
J    in  the  past  two  years  since  the  first 
'  .  Symposium 

• to  provide  a  forum  for  the  exchange  of  ideas 
among  researchers,  manufacturers,  and 
cl inicians 

•to  identify  clinical  problems  which  might  be 
solved  by  ultrasonic  tissue  characterization 

I  •to  identify  research  opportunities 

j  •to  promote  the  transfer  of  new  technological 
'    advances  in  medical  ultrasound  to  commercial 
appl ication 

'  •to  explore  the  potential  for  using  ultrasonic 

tissue  characterization  as  a  mass  screening 
!    technique  for  breast  cancer. 

The  Symposium,  which  extended  over  three  days, 
fas  cosponsored  by  the  National  Bureau  of  Stan- 
llards  (NBS),  the  National  Science  Foundation  (NSF), 
'.nd  the  National  Institutes  of  Health  (NIH).  The 
introductory  Session  opened  with  a  welcoming 
iddress  by  Dr.  E.  Ambler,  Director  of  NBS.  He 
'!poke  of  the  rapid  growth  in  the  application  of 
jiltrasonic  diagnosis.    The  U.S.  market  was  esti- 
mated to  be  $64M  in  1977,  increasing  at  31  percent 
i^er  year.    Ultrasonic  diagnosis  was  apparently 
;afe.    Quantitative  methods  were  being  developed, 
Accelerated  by  the  work  of  the  Tissue  Signature 
'roject  supported  by  NSF.    One  of  the  most  exciting 
liossibilities  was  that  ultrasound  might  be  useful 
iin  mass  screening  for  breast  cancer.    The  interest 
ijind  involvement  of  NBS  in  these  medical  applica- 
'|;ions  was  increasing.    Dr.  A.  J.  Eggers,  Jr., 
'Assistant  Director  for  Research  Applications,  NSF, 
Emphasized  the  importance  of  interagency  collabora- 
ipion.    He  pointed  out  the  dramatic  decline  in  the 
jcost  of  microelectronic  systems;  forecasts  indica- 
jced  that  tasks  requiring  microelectronics  presently 
:bosting  $1M  would  cost  (  1  977  money)  $100K  in  five 
j/ears,  and  only  $10K  in  ten  years.    He  predicted 
jthe  imminent  end  of  the  "visible"  computer  age, 
'jinticipating  the  development  of  displays  with 
J'ntegrated  microelectronics  for  operator  inter- 
action.   He  referred  to  the  1973  findings  of  the 
kudy  group  sponsored  by  NSF,  and  to  the  resulting 
initiation  of  Experiment  No.  5,  which  was  designed 
io  test  the  acceleration  of  technology  transfer  to 
■industry  through  the  vehicle  of  diagnostic  ultra- 
50und.    This  had  been  followed  by  the  preparation 
py  the  Alliance  for  Engineering  in  Medicine  and 
Siology  (AEMB),  with  NSF  support,  of  a  five- 
/ear  research  and  development  agendum;  23  research 
categories  had  been  identified.    Finally,  in 
addition  to  many  project  grants,  NSF  was  supporting 
the  Tissue  Signature  Project  of  the  Carnegie-Mellon 


Institute  of  Research.    Dr.  Eggers  was  followed  by 
Dr.  M.  B.  Lipsett,  Director,  Clinical  Center,  NIH. 
Dr.  Lipsett  outlined  the  NIH  program  of  support  for 
diagnostic  ultrasonic  research,  presently  running 
at  about  $5M  per  year.    The  National  Cancer  Insti- 
tute (NCI)  was  funding  projects  concerned  with 
breast  pathology  imaging  and  with  endoscopy.  The 
National  Heart  and  Lung  Institute  (NHLI)  was 
involved  with  real-time  imaging  and  with  Doppler 
studies  of  blood  flow.    The  National  Institute  of 
General  Medical  Sciences  (NI6MS)  was  supporting 
research  in  imaging  systems  and  the  biological 
effects  of  ultrasound.    The  Clinical  Center  was 
developing  a  real-time  scanner,  and  was  collabora- 
ting with  NBS  in  the  construction  of  a  comprehensive 
tissue  analysis  system.    Dr.  Lipsett  stated  that 
NBS  expected  to  continue  to  be  concerned  with  the 
advancing  horizon  of  ultrasonic  diagnostics. 

The  Symposium  Chairperson,  Dr.  M.  Linzer  (NBS, 
Gai thersburg ,  Maryland)  discussed  the  present 
knowledge  of  the  ultrasonic  properties  of  tissues. 
He  reviewed  the  advantages  of  ultrasonic  diagnosis, 
particularly  in  comparison  with  x-ray  imaging.  He 
demonstrated  these  by  state-of-the-art  illustra- 
tions of  ultrasonic  data  obtained  in  several  leading 
laboratories  --  Ultrasonics  Institute,  Sydney, 
Australia,  Duke  University,  Durham,  North  Carolina, 
and  Horizons  Research  Laboratories,  Fort  Lauderdale, 
Florida.    Comparably  impressive  results  were  begin- 
ning to  be  obtainable  with  commercial  instruments. 
Dr.  Linzer  went  on  to  discuss  basic  data  necessary 
for  further  development  of  ultrasonic  diagnostics 
and  he  spoke  of  new  methods  of  display.    He  then 
recalled  the  First  International  Symposium  on 
Ultrasonic  Tissue  Characterization  which  had  been 
held  two  years  previously  at  NBS  in  Gai thersburg . 
That  meeting  had  been  attended  by  more  than  200 
people  and  the  proceedings  were  recorded  in  NBS 
Special  Publication  453.    As  a  result  of  the 
meeting,  NSF  had  supported  the  Ultrasonic  Tissue 
Signature  Working  Group  and  the  Advisory  Council 
of  Users.    Data  on  tissue  parameters  were  being 
updated,  and  libraries  of  algorithms  and  scans 
were  to  be  collected.    In  the  present  Symposium, 
there  were  to  be  more  than  50  papers,  both  from 
within  the  United  States  and  by  15  speakers  from 
nine  countries  outside  the  United  States.    It  was 
evident  that  measurements  were  more  difficult  to 
make  in  the  clinic  than  in  the  laboratory.  Dr. 
Linzer  outlined  the  conference  program  which  was 
designed  to  focus  sharply  on  measurement  of  tissue 
characteristics.    He  emphasized  that  other  consid- 
erations were  also  vital  for  clinical  success,  and 
that  consequently  the  program  for  the  Third  Sympo- 
sium, which  would  be  held  in  June  1978,  would  have 
an  expanded  scope  and  would  include  imaging  and 
Doppler  techniques. 

The  first  scientific  session.  Velocity  and 
Attenuation  I:  Mechanisms,  began  with  R.  L.  Johnston, 
S.  A.  Goss,  V.  Maynard,  J.  K.  Brady,  L.  A.  Frizzell, 
W.  D.  O'Brien,  Jr.,  and  F.  Dunn  (University  of 
Illinois,  Urbana,  Illinois)  reviewing  ultrasonic 


3 


propagation  properties.    Different  materials  were 
considered  in  order  of  increasing  complexity.  In 
water,  attenuation  was  proportional  to  the  square 
of  the  frequency,  although  the  absolute  value  of 
attenuation  was  rather  greater  than  that  which 
would  be  expected  to  be  due  to  viscosity.  Solu- 
tions of  amino  acids  had  similar  properties  to 
those  of  water;  while  those  of  polypeptides 
exhibited  greater  absorption  probably  due  to  such 
interactions  as  helix-coil  rearrangements  and  pro- 
ton transfers.    Proteins  were  made  up  of  large 
molecules,  with  relaxational  absorption.  Tissues 
had  characteristic  attenuation  dependency  on 
temperature  and  frequency.    In  general,  both 
increasing  attenuation  and  velocity  seemed  to  be 
associated  with  decreasing  water  content  and 
with  increasing  protein  and  collagen  content.  Bone 
and  lung  exhibited  more  complicated  relationships. 
For  non-gas-containing  soft  tissues,  however, 
tissues  with  similar  physiological  functions  had 
similar  attenuation,  and  this  property  might  be 
used  to  characterize  tissues.    In  discussion,  the 
authors  stated  that  the  strong  frequency  dependence 
of  velocity  in  lung  might  be  due  to  changing  ultra- 
sonic pathways,  and  they  agreed  that  the  role  of 
collagen  deserved  further  study.    E.  A.  Carstensen 
(University  of  Rochester,  Rochester,  New  York)  then 
reviewed  the  absorption  of  ultrasound  in  tissue. 
He  considered  first  the  relatively  small  contribu- 
tion to  total  absorption  due  to  the  cellular  struc- 
tural components  of  tissue.    The  linear  frequency 
dependence  of  absorption  suggested  a  spectrum  of 
relaxation  processes.    The  temperature  coefficient 
of  attenuation  had  been  measured  for  few  tissues, 
and  data  on  dispersion  were  sparse.  Experimental 
results,  however,  were  consistent  with  the  theory. 
Mechanisms  of  ultrasonic  attenuation  in  soft  tissue 
were  further  discussed  by  M.  O'Donnell  and  J.  G. 
Miller  (Washington  University,  St.  Louis,  Missouri). 
They  presented  a  theoretical  analysis  of  losses  due 
to  cellular  motion,  and  showed  that  in  the  limit 
for  small  scatterers  there  was  a  linear  frequency 
dependence  of  attenuation.    They  concluded  that 
losses  due  to  microscopic  inhomogeneities  accounted 
for  a  substantial  fraction  of  the  total  observed 
attenuation  in  soft  tissues.    The  authors  were 
questioned  about  the  effects  of  concentration, 
which  they  said  they  had  taken  into  account,  and 
about  the  existence  of  dispersion,  which  they 
believed  to  be  due  to  relaxation. 

Session  2,  Scattering  and  Attenuation,  dealt 
with  techniques  for  estimating  attenuation  from 
measurements  of  scattering.    F.  L.  Lizzi  and  M. 
Elbaum  (Riverside  Research  Institute,  New  York, 
New  York)  described  clinical  spectrum  analysis 
techniques  for  tissue  characterization.    They  had 
measured  the  echo  amplitude  decrement  from  within 
a  narrow  gated  sample  tracked  increasingly  deeply 
into  homogeneous  tissue  layers.    Moreover,  spec- 
tral analysis  revealed  characteristic  periodici- 
ties for  different  histologies.    Within  the  eye 
and  orbit,  retina,  hemorrhage,  fat,  melanoma  and 
glioma  could  be  distinguished.    Using  a  model 
embodying  scattering  (from  a  fixed  interface)  and 
attenuation,  S.  Levi  and  J.  Keuwez  (Hopital 
Uni versitaire  Brugmann,  Brussels,  Belgium)  re- 
ported an  attempt  to  characterize  tissues  in  vivo 
by  differential  attenuation  measurements  on  the 
basis  of  observations  made  at  two  frequencies 
(2  and  4  MHz).    Results  obtained  in  vivo  with 
leiomyoma  and  cyst  were  encouraging^    RT  Kuc  and 


M.  Schwartz  (Columbia  University,  New  York,  New 
York)  and  N.  Finby  and  F.  Dain  (St.  Luke's  Hospital 
Center,  New  York,  New  York),  in  a  joint  paper  on 
statistical  estimation  of  the  acoustic  attenuation 
coefficient  slope  for  liver  tissue  from  reflected 
ultrasonic  signals,  discussed  the  separation  of 
fixed  and  refrigerated  liver  tissue  on  the  basis 
of  spectral  differences  in  the  backscattered  signals. 
The  liver  was  treated  as  a  random  linear  filter. 

This  Session  was  closely  related  to  Session  3, 
Scattering.    The  surface  of  the  lung  was  considered 
by  T.  L.  Rhyne  (Massachusetts  Institute  of  Tech- 
nology, Cambridge,  Massachusetts)  to  have  an  ultra- 
sonic tissue  signature  which  could  be  represented 
by  a  stochastic  model.    Dr.  Rhyne  took  special 
care  to  measure  the  characteristics  of  the  ultra- 
sonic transmitter,  transducer,  and  receiver  which 
he  used.    Recent  developments  in  research  into 
angle-scan  and  frequency-swept  ultrasonic  scatter- 
ing characterization  of  tissue  were  reported  by 
R.  C.  Waag,  P.  P.  K.  Lee,  R.  M.  Lerner,  L.  P. 
Hunter,  R.  Gramiak,  and  E.  A.  Schenk  (University 
of  Rochester,  Rochester,  New  York).    Using  an 
experimental  arrangement  giving  a  true  analog  of 
Bragg  scattering,  they  demonstrated  distinctive 
correlations  for  normal,  cirrhotic,  and  fatty 
liver.    Measurements  of  tissue  in  vitro  were 
related  to  models  and  measurements  of  known  suspen- 
sions.   Some  of  the  difficulties  of  measuring 
scattering  had  been  solved  by  F.  E.  Barber  (Harvard 
Medical  School,  Boston,  Massachusetts),  who 
described  ultrasonic  microprobe  methods  for  tissue 
characterization.    Dr.  Barber's  microprobe  operated 
at  10  MHz,  and  had  an  almost  Gaussian  focus  with 
a  diameter  of  about  0.2  mm.    Where  large  inter- 
faces existed,  they  dominated  the  scattering.  The 
relationships  between  interface  dimensions  and 
scattering  directivity  were  elegantly  demonstrated 
by  reference  to  a  micrograph  of  the  experimental 
tissue.    J.  M.  Reid  and  K.  K.  Shung  (Providence 
Medical  Center  and  Institute  of  Applied  Physiology 
and  Medicine,  Seattle,  Washington)  presented  quan- 
titative measurements  of  scattering  by  heart  and 
liver.    They  had  plotted  the  scattering  cross- 
sections  of  12-20  cm'  samples  of  tissue  for  many 
orientations,  and  they  discussed  these  in  relation 
to  scattering  equations  for  isotropic,  plane,  and 
lossy  scatterers.    In  particular,  they  considered 
the  effects  of  attenuation,  both  within  the  sample 
and  in  the  region  between  the  transducer  and  the 
sample.    The  dependence  of  ultrasound  backscatter 
from  human  liver  tissue  on  frequency  and  protein/ 
lipid  composition  was  discussed  by  M.  Freese  and 
E.  A.  Lyons  (Radionics  Limited,  Montreal,  Quebec, 
Canada).    They  had  used  a  system  with  a  bandwidth 
equal  to  25  percent  of  the  center  frequency,  and 
found  that  a  frequency  of  2.25  MHz  gave  the  best 
separation  between  normal  and  fatty  liver.  Signi- 
ficant correlations  of  the  backscatter  with  protein 
content  were  observed  in  normal  liver  and  with  both 
protein  and  lipid  content  in  abnormal  fatty  liver. 
Final  lyjn  this  Session,  M.  Hanss  and  M.  Boynard, 
(Faculte  de  Medecine  et  de  Biologie,  Bobigny, 
France)  reported  data  on  ultrasound  backscattering 
from  blood  and  its  hematocrit  and  erythrocyte 
aggregation  dependence.    They  stated  that  fluctua- 
tions in  backscattered  amptitude  were  increased  in 
frequency  with  higher  sedimentation  rates  during 
the  sedimentation  process,  and  explained  this  by 
a  theory  involving  the  existence  of  rouleaux  in 
Kigh  sedimentation  rate  blood,  and  the  occupancy 


4 


of  potential  scattering  sites.    They  suggested 

that  eventually  a  noninvasive  ultrasonic  measure- 
ment might  supersede  the  contemporary  pathology 
test. 

Session  4,  Tumor  Doppler  Signatures,  consis- 
ted of  two  papers  describing  measurements  of 
blood  flow  changes  due  to  malignancy.    P.  N.  T. 
Wells,  M.  Halliwell,  R.  A.  Montford,  R.  Skidmore, 
A.  J.  Webb,  and  J.  P.  Woodcock,  (Bristol  General 
Hospital  and  Bristol  Royal  Infirmary,  Bristol, 
United  Kingdom)  explained  the  hypothesis  that  an 
in  situ  carcinoma  only  becomes  a  rapidly  prolif- 
erating tumor  after  initiation  of  vascularization. 
They  showed  that  asymmetrical  arterial  blood  flow 
patterns  might  be  found  with  two  breasts,  one 
containing  a  malignant  tumor.    This  was  confirmed 
in  the  following  paper  on  Doppler  echography  by 
G.  Dale,  Ch.  M.  Gros,  M.  Gautherie,  and  B.  Gairard 
(Senolgie-Hospices  Civils,  Strasbourg,  France). 
These  authors  had  used  a  pulsed  8  MHz  Doppler 
system  to  study  arterial  flow  patterns,  and  they 
explained  the  complementary  nature  of  thermography 
in  imaging  venous  flow.    Returning  to  the  paper  by 
Wells  and  his  colleagues,  these  authors  had  gone 
on  to  describe  the  discovery  with  an  8  MHz  contin- 
uous wave  system  of  abnormal  flow  signals,  appar- 
ently associated  with  malignant  tumor  neovascular- 
ization, arising  from  within  the  breast  lesions 
themselves.    Such  signals  are  not  found  in  normal 
breasts,  or  associated  with  benign  tumors.  They 
reported  that  they  had  also  detected  similarly 
augmented  blood  flow  signals,  using  a  2  MHz  pulsed 
Doppler  instrument,  from  within  a  malignant  pan- 
creatic tumor.    In  discussion,  it  was  explained 
that  the  breast  tumor  signals  were  sometimes 
strongest  at  the  edge  of  the  lesion,  perhaps 
because  of  necrosis  at  the  center. 

Three  papers  reporting  studies  of  cardiac 
muscle  were  presented  in  Session  5,  Attenuation 
and  Velocity  II:    Myocardium.    T.  D.  Franklin,  Jr., 
N.  T.  Sanghvi ,  F.  J.  Fry,  K.  M.  Egenes,  and  A.  E. 
Weyman  (Indiana  School  of  Medicine  and  Indianapolis 
Center  for  Advanced  Research,  Indianapolis,  Indiana) 
described  ultrasonic  tissue  characterization  studies 
of  ischemic  and  infarcted  myocardium.    The  impetus 
for  this  research  came  from  the  non-specificity 
of  conventional  echocardiography  in  evaluating 
infarction.    Their  preliminary  results  showed  that 
attenuation  fell  immediately  after  infarct.    A  more 
extensive  study  by  J.  G.  Miller,  M.  O'Donnell, 
J.  W.  Mimbs,  and  B.  E.  Sobel  (Washington  University, 
St.  Louis,  Missouri)  of  ultrasonic  attenuation  in 
normal  and  ischemic  myocardium,  involved  measure- 
ments made  with  a  cadmium  sulfide  phase-insensi- 
tive receiving  transducer.    The  range  of  attenua- 
tion variation  in  normal  myocardium  was  about  12 
to  15  percent,  and  attenuation  fell  by  about  20 
percent  when  the  temperature  of  the  sample  was 
increased  from  20°  to  37  °C.    The  slope  of  a 
least-squares  line  fitted  to  the  attenuation 
coefficient  versus  frequency  data  served  as  a  con- 
venient ultrasonic  index.    In  dogs  with  experi- 
mentally-induced ischemia,  this  slope  was  less 
than  the  normal  value  if  sacrifice  was  within 
about  1  day  after  occlusion,  but  with  later 
killing  it  became  greater.    This  may  reflect  an 
increase  in  collagen  content  in  necrotic  scar 
tissue.    Acoustic  microscopic  analysis  of  myocar- 
dium by  D.  E.  Yuhas  and  L.  W.  Kessler  (Sonoscan 
Inc.,  Bensenville,  Illinois)  had  yielded  100  MHz 


data  on  attenuation  and  velocity  in  formalin- 
fixed  tissue.    In  kidney,  it  had  previously  been 
shown  that  the  effect  on  attenuation  of  formalin 
fixing  may  not  be  significant.    In  normal  myocar- 
dium, the  attenuation  was  about  3.5  to  5.7  dB  cm"^ 
MHz-^.    Moreover,  the  detailed  demonstration  at 
high  frequency  of  structures  which  were  of  wave- 
length-order dimensions  at  low  megahertz 
frequencies  was  directly  relevant  to  the  study 
of  scattering  in  conventional  echography. 

The  second  day  of  the  Symposium  began  with 
Session  6,  Image  Reconstruction.    G.  H.  Glover 
(General  Electric  Co.,  Milwaukee,  Wisconsin) 
discussed  in  vivo  measurement  of  ultrasonic  refrac- 
tive index  distributions  in  human  breasts  by  time- 
of-f light  tomography.    His  laboratory  instrument 
was  a  5  MHz  transmission  computerized  tomograph. 
The  patient  lay  prone  on  a  canvas  sling  with  one  of 
her  breasts  immersed  in  water  at  32°  to  36  °C. 
Nine  minutes  were  required  to  scan  a  slice  of  5  mm 
thickness.    Reconstructed  two-dimensional  images 
revealed  cysts  and  fibroadenomas  (having  velocities 
around  2  percent  higher  than  water)  and  malignant 
tumors  (having  velocities  about  4  to  5.5  percent 
lower).    Young  normals  (age  range  24  to  29  years) 
had  quite  wide  velocity  variations,  but  they 
might  not  need  to  be  scanned  in  a  mass  screening 
program.    Histograms  of  pixel  velocity  distribu- 
tions, and  cumulative  pixel  velocity  plots,  might 
provide  accurate  separation  of  normals  from 
patients  with  breasts  made  asymmetrical  by  the 
presence  of  malignancy.    Dr.  Glover  stated  that  he 
was  developing  a  fan  beam  detector  array  with  127 
elements,  designed  to  reduce  the  examination  time 
to  10  seconds  per  breast.    Questions  were  raised 
by  the  audience  concerning  path  straightness  (Dr. 
Glover  had  ray  plots),  resolution  (which  was  1  ns, 
although  10  ns  would  have  been  adequate),  point 
spread  function  (MTF  -  1  mm),  and  accessibility 
(slices  could  be  within  20  mm  of  the  chest  wall 
in  the  experimental  system).    Using  a  synthetic 
"fan  beam,"  R.  Bal asubramani an ,  J.  F.  Greenleaf, 
P.  J.  Thomas,  and  S.  A.  Johnson  (Mayo  Clinic, 
Rochester,  Minnesota)  reported  measurements  of 
temperature  coefficients  of  ultrasonic  speed  in 
various  human  tissue.    Apart  from  that  of  fat,  the 
temperature  coefficients  of  liver,  kidney,  and 
other  tissues  were  all  about  2  m  s"^K"^.  The 
possibility  that  tumors  might  be  identifiable 
because  of  their  elevated  temperatures  was  men- 
tioned during  the  discussion,  but  the  authors  had 
not  investigated  velocity  changes  with  temperature 
in  tumor  tissue.    Results  obtained  with  computed 
tomographic  reconstruction  of  attenuation  images 
had  previously  been  disappointing,  but  new 
approaches  were  suggested  by  A.  C.  Kak  and  K.  A. 
Dines  (Purdue  University,  West  Lafayette,  Indiana) 
in  their  paper  on  signal  processing  for  the 
measurement  of  attenuation  of  soft  tissue.  They 
pointed  out  that  reflection  contributed  a  loss  of 
about  4  percent  at  each  major  interface,  making 
algebraic  reconstruction  techniques  unsatisfactory. 
Other  frequency  and  time  domain  methods,  such  as 
polynomial  fitting,  frequency  averaging  and 
energy  analysis,  might  prove  to  be  good  candidates 
for  attenuation  measurements.    For  example,  a 
reconstruction  of  a  section  through  a  dog's 
heart  by  the  energy  method  demonstrated  immunity 
to  reflection  artifacts.    In  discussion,  members 
of  the  audience  emphasized  the  importance  of 
clinical  practicability.    S.  A.  Johnson,  J.  F. 


5 


Greenleaf,  and  B.  Rajagopalan  (Mayo  Clinic, 
Rochester,  Minnesota)  and  R.  C.  Bahn  and  B. 
Baxter  (University  of  Utah,  Salt  Lake  City,  Utah), 
in  a  joint  paper  on  the  future  role  of  high 
spatial  resolution  ultrasonic  measurement  tech- 
niques for  characterization  of  static  and  moving 
tissue,  discussed  the  determination  by  computer- 
ized tomography  of  three-dimensional  fluid  flow 
and  temperature.    They  proposed  the  use  of 
seismological  approaches  for  high-resolution 
imaging.    Another  new  method,  possibly  leading 
to  mapping  true  ultrasonic  backscatter  and 
attenuation  distributions  in  tissue  --  a  digital 
reconstruction  approach  --  was  suggested  by  F.  A. 
Duck  and  C.  R.  Hill  (Institute  of  Cancer  Research, 
Sutton,  United  Kingdom).    By  assuming  that 
attenuation  and  scattering  were  constant  and  iso- 
tropic within  each  volume  element  of  the  object, 
they  showed  that  the  corresponding  pixel  values  in 
the  image  could  be  calculated  by  algebraic  recon- 
struction techniques  (ART)  in  about  six  iterations. 
Discussion  revealed  that  isotropic  assumptions 
might  not  be  satisfied  in  clinical  practice. 

Sessions  7  and  8,  Signal  Processing  and  Pattern 
Recognition  I  and  II,  contained  a  total  of  twelve 
papers.    Region-of-interest  analysis  methods  were 
used  in  feature  extraction  techniques  designed  to 
improve  recognition  of  patterns  in  ultrasonic 
pictures  of  the  prostate  for  tumor  diagnosis,  by 
W.  V.  Seelen  (Johannes  Gutenberg  Universitat, 
Mainz),  E.  G.  Lock  and  G.  Wessels  (Deutsche  Klinik 
fiir  Diagnostik,  Wiesbaden),  U.  Scheiding  (Battelle 
Institut,  Frankfurt),  and  A.  Gaca  (Deutsche  Klinik 
fiir  Diagnostik,  Wiesbaden,  Federal  Republic  of 
Germany).    J.  C.  Birnholz  (Harvard  University, 
Boston,  Massachusetts)  then  reviewed  the  elements 
of  visual  pattern  recognition  in  ultrasonography. 
He  showed  some  remarkably  excellent  grey-scale 
pictures.    The  influence  of  signal  processing,  and 
particularly  of  the  detection  method,  on  the  quality 
of  echograms  was  discussed  by  I.  Beretsky,  D.  Arnold, 
and  J.  Cason  (Searle  Ultrasound  Research  and  Ad- 
vanced Development,  Spring  Valley,  New  York),  in 
their  paper  on  impulse  detection  in  pulse  echo 
ultrasound  --  recent  in  vitro  experiments  with  a 
human  aorta.    This  was  followed  by  a  theoretical 
analysis  of  instantaneous  power  spectra  as  applied 
to  spectra-color  ultrasonography,  by  W.  D.  Jennings, 
E.  Holasek,  and  E.  W.  Purnell  (Case  Western  Reserve 
University,  Cleveland,  Ohio).    The  method  was 
based  on  the  separate  display  in  the  primary  colors 
on  the  same  image  of  three  frequency  bands  consti- 
tuting the  echo  signals.    Session  7  ended  with  a 
paper  by  C.  K.  Kuni  (University  of  Colorado 
Medical  Center,  Denver,  Colorado)  on  tissue  iden- 
tification by  spectral  analysis  of  scattered  ultra- 
sound.   Dr.  Kuni  discussed  some  of  the  problems  of 
obtaining  and  interpreting  backscattered  spectra. 

The  following  Session  8  began  with  a  descrip- 
tion of  a  comprehensive  ultrasonic  tissue  analysis 
system  by  M.  Linzer,  S.  I.  Parks,  R.  W.  Shideler, 
S.  J.  Norton,  F.  P.  Higgins,  and  D.  R.  Dietz 
(National  Bureau  of  Standards,  Washington,  D.C.) 
and  J.  L.  Doppman  and  T.  H.  Shawker  (National 
Institutes  of  Health,  Bethesda,  Maryland).  Dr. 
Linzer  described  a  dynamically-focused  system 
using  an  expanding-aperture  annular  array  with  an 
approximately  constant  F-number,  and  an  rf  sub- 
system capable  of  operating  with  impulses,  gated 
continuous  wave,  and  chirp  pulses.    Chirp  operation 


with  8:1  compression  had  been  achieved.  An 
ultrafast  signal  averager  was  developed  for  im- 
provement of  the  signal-to-noise  ratio  of  A-scans. 
This  instrument  had  a  sampling  rate  of  50  MHz  and 
a  4-bit  analog-to-digital  converter,  and  had 
potential  applications  in  the  examination  of  obese 
patients,  the  use  of  higher  frequencies,  and  the 
use  of  inefficient  transducers.    Next,  an  instru- 
ment called  the  "SonoChromascope"  was  described. 
Interfaced  to  a  commercial  two-dimensional  B- 
scanner,  this  device  formed  a  versatile  digital 
real-time  acquisition,  signal  processing,  and 
display  system.    The  250  kbyte  memory  allowed 
operation  in  normalized  averaging,  maximum, 
minimum,  and  parameter  comparison  modes.  Gray-scale, 
color,  and  window  displays,  as  well  as  digital 
readout  of  area  and  average  echo  amplitude  in 
regions-of-interest  were  possible.  Finally, 
Dr.  Linzer  stated  that  other  methods,  including  CT 
scanning,  were  being  investigated  at  NBS.  There 
was  discussion  about  the  rather  low  digitization 
accuracy  of  commercial  transient  recorders  which 
do  not  contain  sample-and-hold  circuits.  The 
subject  of  spectrum  analysis  was  again  considered 
in  the  following  paper  by  L.  Joynt,  D.  Boyle,  H. 
Rakowski,  and  W.  Beaver  (Stanford  University, 
Stanford,  California),  on  the  identification  of 
tissue  parameters  by  digital  processing  of  real- 
time ultrasonic  clinical  data.    Data  obtained  with 
a  real-time  phased  array  sector  scanner  from 
selected  regions  within  the  myocardium  in  various 
clinical  conditions  were  subjected  to  spectral 
analysis.    Although  there  was  substantial  overlap 
between  groups,  spectra  from  patients  with  myocar- 
dial infarcts  had  much  less  fluctuation  in  their 
means  and  variances  as  a  function  of  time.    J.  C. 
Gore  and  S.  Leeman  (Royal  Postgraduate  Medical 
School,  London,  United  Kingdom)  presented  two  short 
papers.    The  first,  on  the  theoretical  evaluation 
of  backscatter  of  ultrasonic  pulses  from  human 
tissue  and  its  implications  for  tissue  characteri- 
zation, dealt  with  the  effects  of  the  characteris- 
tics of  pulser,  transducer,  and  receiver.  The 
second  paper,  on  autocorrelation  analysis  of  A- 
scans  in  vivo  and  the  possible  clinical  application 
of  temporal  changes  in  echo  characteristics, 
revealed  similarities  between  the  echoes  from 
different  specimens  of  the  same  types  of  tissue, 
while  different  types  of  tissue  had  dissimilar 
autocorrelation  functions.    With  the  goal  of  imple- 
menting computer  spectral  analysis  of  ultrasonic 
A-mode  echoes,  D.  E.  Robinson  (Ultrasonics  Institute, 
Sydney,  Australia)  had  developed  an  on-line  data 
acquisition  system  based  on  a  clinical  scanner, 
10  MHz  sample  rate  transient  recorder,  and  Inter- 
data  85  computer.    Special  consideration  had  been 
given  by  K.  Preston,  Jr.  (Carnegie-Mellon  Univer- 
sity, Pittsburgh,  Pennsylvania)  and  M.  L.  Skolnik 
(University  of  Pittsburgh,  Pittsburgh,  Pennsylvania) 
to  the  application  of  pattern  recognition  techniques 
to  the  separation  of  ultrasonic  echoes  from  differ- 
ent types  of  tissue.    Digitized  A-scans  had  been 
converted  to  sound  patterns  as  a  preliminary  to 
testing  the  feasibility  of  using  the  ear  to  dis- 
tinguish the  sound  patterns  from  different  types 
of  tissue.    The  A-scans  had  been  analyzed  in  terms 
of  skew,  kurtosis,  and  other  statistical  parameters. 
Initial  results  had  been  obtained  from  normal  and 
neoplastic  kidney.    In  order  to  determine  the 
correlation  function  of  echoes  from  a  region-of- 
interest  identified  on  a  two-dimensional  B-scan, 
J.  Eraser  and  G.  S.  Kino  (Stanford  University, 


6 


Stanford,  California)  and  J.  Birnholz  (Harvard 
Medical  School,  Boston,  Massachusetts)  reported 
the  use  of  cepstrum  processing  for  tissue  analysis. 
The  transducer  response  appeared  mainly  centered 
at  the  cepstrum  origin,  and  it  was  gated  out  so 
that  the  inverted  cepstrum  allowed  the  well-decon- 
voluted  autocorrelation  response  of  the  tissue 
alone  to  be  observed.    Finally  in  this  lengthy 
pair  of  sessions,  R.  D.  Lepper,  R.  Reuter,  and 
H.  G.  Trier  (Universitat  Bonn,  Bonn,  Federal 
Republic  of  Germany)  pointed  out  the  advantage  of 
asychronous  writing  and  reading  on  a  storage  tube 
in  time-stretching  ophthalmic  A-scans  in  order  to 
improve  the  precision  of  digitization. 

Session  9  was  a  Panel  Discussion  on  Ul tra- 
sonic  Diagnosis  of  Breast  Cancer.    The  cochair- 
person,  P.  N.  T.  Wells  (Bristol  General  Hospital, 
Bristol,  United  Kingdom)  in  his  introductory  remarks 
pointed  out  that  mortality  from  breast  cancer  had 
not  improved  over  the  last  20  years.    He  explained 
that  in  the  Western  World,  breast  cancer,  which  had 
many  different  pathological  types,  killed  1  in  50 
women,  was  the  main  cause  of  death  in  females  in 
the  age  range  40  to  44,  and  in  the  U.S.  had  an 
annual  economic  cost  of  $200M.    Existing  diagnostic 
methods  were  inadequate  for  mass  screening.  Interest 
in  ultrasonic  techniques  hinged  around  the  possi- 
bility that  earlier  detection  by  screening  might 
improve  prognosis.    Cochai rperson  W.  Pomerance 
(National  Institutes  of  Health,  Bethesda,  Maryland), 
defined  screening  as  the  application  of  known 
methods  to  an  asymptomatic  population  to  detect 
early  cancer.    He  considered  that  a  successful 
method  for  breast  screening  would  need  to  detect 
tumors  of  less  than  5  mm  diameter,  and  that  it 
should  be  free  from  hazard  and  have  at  least  a  90 
percent  success  rate.    Encouraging  results  with 
high-resolution  scanning  led  L.  Weiss  (Roswell 
Park  Memorial  Institute,  Buffalo,  New  York)  to 
suggest  that  ultrasound  might  be  capable  of  detec- 
ting lesions  of  2  mm  diameter.    This  would  be  an 
important  although  numerically  modest  improvement 
over  the  present  ability  to  detect  5  mm  lesions  by 
manual  palpation.    He  explained  that  it  had  been 
estimated  that  screening  the  whole  adult  female 
population  with  an  effective  ultrasonic  method,  if 
one  should  be  developed,  would  have  cost  $620M  per 
year  in  1971,  and  he  felt  that  it  would  therefore 
be  necessary  to  restrict  screening  to  women  known 
to  be  at  risk,  or  over  50  years  old.    J.  L. 
Doppman  (National  Institutes  of  Health,  Bethesda, 
Maryland)  described  the  role  of  mammography,  and 
stated  that  the  x-ray  exposure  was  presently  less 
than  0.01  Gy.    Mammography  could  ensure  that  the 
lesion  was  removed  by  the  surgeon.    Five  years  of 
experience  of  ultrasonic  visualization  of  the 
breast  in  Japan  were  reviewed  by  T.  Kobayashi 
(National  Cancer  Center  Hospital,  Tokyo,  Japan). 
Ultrasonic  screening  was  being  made  available  by 
means  of  mobile  units.    G.  Dale  (Senologie-Hospices 
Civils,  Strasbourg,  France)  described  the  comple- 
mentary use  of  ultrasonic  scanning,  both  pulse- 
echo  and  Doppler,  together  with  thermography  and 
mammography,  to  decrease  the  overall  error  rate. 
The  results  of  research  into  high  resolution  pulse- 
echo  imaging  of  the  breast,  and  the  problems  of 
analyzing  the  scans,  were  presented  by  G.  Baum 
(Albert  Einstein  College  of  Medicine,  New  York, 
New  York).    Elizabeth  K.  Fry  (Indiana  University 
School  of  Medicine  and  Indiana  University  Hospital, 
Indianapolis,  Indiana)  described  a  versatile  ultra- 


sonic breast  scanning  research  instrument.  She 
presented  some  scans,  and  discussed  their  interpre- 
tation, emphasized  the  importance  of  shadows,  and 
the  different  patterns  obtained  with  different 
transducers  and  frequencies.    Cautious  optimism 
was  expressed  by  D.  E.  Robinson  (Ultrasonics 
Institute,  Sydney,  Australia).    He  demonstrated 
some  excellent  scans,  obtained  with  a  4  MHz 
focused  beam  in  a  water  bath  scanner,  and  spoke 
of  eight  features  for  which  the  scans  required  to 
be  examined  in  reaching  a  diagnosis.    J.  F. 
Greenleaf  (Mayo  Clinic,  Rochester,  Minnesota) 
described  the  role  of  ultrasonic  computed  tomog- 
raphy in  obtaining  data  on  tissue  properties  and 
in  correcting  two-dimensional  B-scans.    G.  H. 
Glover  (General  Electric  Company,  Milwaukee, 
Wisconsin)  had  previously  talked  about  his  work 
on  ultrasonic  computed  tomography  in  a  paper  pre- 
sented during  Session  6.    As  time  was  short.  Dr. 
Glover  did  not  add  to  the  information  which  he  had 
already  given  concerning  the  increase  in  velocity 
associated  with  malignancy  in  breast  tissue.  For 
the  same  reason.  Dr.  Wells  did  not  elaborate  on 
the  potential  of  neovascularization  blood  flow 
Doppler  detection,  as  he  had  mentioned  this  in 
Session  4.    M.  Linzer  (National  Bureau  of  Standards, 
Washington,  D.C.)  emphasized  the  vast  quantity  of 
data  which  would  be  obtained  in  scanning  only  one 
patient,  and  of  the  refinements  in  methodology 
which  would  be  necessary  to  make  ultrasonic  breast 
screening  feasible.    Members  of  the  Panel  and  the 
audience  then  began  an  open  discussion,  and  several 
points  were  made.    The  demonstration  of  microcalci- 
fication  in  ultrasonic  scans  was  possible  if 
sufficiently  extensive.    Ultrasonic  scanning  was 
beneficial  in  young  women  with  "lumpy"  breasts. 
Mammography  did  not  contribute  further  when  ultra- 
sound had  already  detected  a  lesion.  Present 
commercial  equipment  was  unsuitable  for  starting  a 
mass  screening  program.    Mammography  would  in 
principle  never  be  an  acceptable  mass  screening 
method,  as  it  would  always  involve  exposure  to 
some  radiation.    Dr.  Wells  closed  the  discussion 
with  a  short  summary.    Mass  screening  of  selected 
groups  of  the  female  population  would  be  worth- 
while if  a  suitable  technique  could  be  developed. 
Such  a  technique  might  well  be  based  on  ultrasound 
but  it  would  need  to  be  accurate  and  economical. 
The  present  lines  of  research  were  promising; 
more  resources  should  be  devoted  to  solving  the 
fundamental  and  technological  problems,  so  that 
earlier  clinical  application  would  be  possible. 

Propagation  Through  Bone  and  Skull  was  the 
subject  of  Session  10,  which  opened  the  final  day 
of  the  Symposium.    A  theory  relating  sonic  velocity 
to  mineral  content  in  bone  was  presented  by  S.  Lees 
(Forsyth  Dental  Center,  Boston,  Massachusetts). 
Bone  had  a  higher  velocity  than  that  predicted  on 
the  basis  of  its  longitudinal  elastic  modulus. 
Dr.  Lees  suggested  that  this  was  due  to  the  bone 
consisting  of  mineral  hydroxyapati te  crystallites 
embedded  in  a  matrix  of  collagen,  behaving  in  the 
same  way  as  a  plastic  filled  with  powdered  mineral. 
H.  S.  Yoon  and  J.  L.  Katz  (Rensselaer  Polytechnic 
Institute,  Troy,  New  York)  considered  the  ultra- 
sonic properties  and  microtexture  of  human  cortical 
bone,  and  showed  that  bone  could  be  treated  as  an 
elastic  dielectric.    The  attenuation  of  ultrasound 
in  cancellous  bone  was  the  subject  of  a  paper  by 
J.  E.  Barger  (Bolt  Beranek  and  Newman,  Cambridge, 
Massachusetts).    He  distinguished  between  phase 


7 


and  group  velocities,  and  showed  that  scattering 
in  the  dipole  made  an  important  contribution  to 
losses  at  frequencies  in  the  range  0.5  to  2  MHz. 
The  results  of  measurements  of  transmission  through 
skull  were  presented,  in  a  discussion  of  the 
acoustic  characteristics  of  the  skull,  by  D.  N. 
White  (Queens  University,  Kingston,  Ontario, 
Canada).    Dr.  White  explained  that  propagating 
ultrasonic  energy  appeared  to  be  distributed 
between  the  bony  and  the  soft  tissue  elements  of 
cancellous  bone  resulting  in  substantial 
attenuation,  although  it  was  transmitted  with 
relatively  low  loss  through  ivory  bone.  Inhomo- 
geneities  led  to  marked  distortion  of  megahertz 
frequency  beams  transmitted  through  the  skull. 
Because  of  this  difficulty,  F.  J.  Fry  (Indianapolis 
Center  for  Advanced  Research,  Indianapolis,  Indiana) 
had  investigated  the  transkull  transmission  of  axi- 
symmetric  focused  ultrasonic  beams  in  the  0.5  to 
1  MHz  frequency  range,  and  its  implications  for 
brain  tissue  visualization,  interrogation,  and 
therapy.    Beams  at  the  relatively  low  frequency  of 
0.5  MHz  were  not  greatly  distorted  in  travelling 
through  skull,  and  there  seemed  to  be  quite  good 
correlation  between  transkull  ultrasonic  two- 
dimensional  brain  scans  and  x-ray  computed  tomo- 
graphs.   The  potential  for  therapy  resulting  from 
the  relatively  low  attenuation  in  the  skull  was 
demonstrated  by  a  lesion  induced  in  lucite  by  a 
beam  of  ultrasound  which  had  passed  through  part 
of  a  cadaver  skull.    The  feasibility  of  visualizing 
the  brain  was  further  demonstrated  by  some  advances 
in  acoustic  imaging  through  the  skull  which  were 
reported  in  a  joint  paper  by  S.  W.  Smith  (Bureau 
of  Radiological  Health,  Rockville,  Maryland),  D.  J. 
Phillips  (University  of  Washington,  Seattle,  Wash- 
ington), and  0.  T.  von  Ramm  and  F.  L.  Thurstone 
(Duke  University,  Durham,  North  Carolina).  Remark- 
able real-time  pictures,  showing  brain  structures 
and  pulsating  arteries,  had  been  made  with  a 
phased-array  sector  scanner.    Treating  the  skull 
as  a  random  acoustic  lens,  optimal  results  were 
theoretically  obtained  at  a  frequency  of  about 
1  MHz  and  an  aperture  width  of  about  35  to  40  mm. 
The  ultrasonic  images  were  well  correlated  with 
pictures  from  a  brain  atlas.    The  possibility  was 
pointed  out  that  phase  correction  for  the  effect 
of  the  skull  could  be  obtained  element-by-element 
across  the  array,  if  a  suitable  intracranial  source 
of  uniform  wavefronts  could  be  devised. 

The  following  Sessions  11  and  12,  Attenuation 
and  Velocity  III  and  IV  had  ten  papers.    S.  A.  Goss, 
R.  L.  Johnston,  V.  Maynard,  L.  Nider,  L.  A.  Frizzell, 
W.  D.  O'Brien,  Jr.,  and  F.  Dunn  (University  of 
Illinois,  Urbana,  Illinois)  reviewed  ultrasonic 
propagation  parameter  measurements.    They  discussed 
the  advantages,  disadvantages,  and  accuracies  of 
resonant  cavities,  interferometers,  and  thermo- 
couple probes,  and  techniques  based  on  velocity 
difference  measurement,  pulse  superposition,  radia- 
tion pressure,  and  time-of-f 1 ight  measurement. 
The  temperature  dependence  of  the  velocity  of  sound 
in  soft  tissues,  a  parameter  of  great  importance  in 
hyperthermia  therapy  for  control  of  cancer,  was 
reported  by  T.  Bowen,  W.  G.  Connor,  R.  L.  Nasoni, 
A.  E.  Pifer,  and  R.  R.  Sholes  (University  of 
Arizona,  Tucson,  Arizona).    With  the  exception  of 
fat,  vegetable  oil,  and  water,  which  had  negative 
temperature  coefficients  of  velocity,  all  the  other 
tissues  investigated  --  spleen,  muscle,  liver, 
and  kidney  --  had  positive  temperature  coefficients 


over  the  range  36°  to  44  °C.    A  simple  but  effec- 
tive device  for  measuring  ultrasonic  propagation 
velocity  in  tissue,  both  in  vitro  and,  with 
suitable  anatomy,  in  vivo,  was  described  by  B.  D. 
Sollish  (Weismann  Institute  of  Science,  Rehovat, 
Israel).    The  following  three  papers  were  con- 
cerned with  ultrasonic  studies  of  the  breast.  T. 
Kobayashi  (National  Cancer  Center  Hospital,  Tokyo, 
Japan)  spoke  about  the  correlation  of  attenuation 
in  breast  cancers  with  connective  tissue  content. 
He  showed  that  stronger  shadowing,  associated 
with  greater  attenuation,  was  correlated  with 
higher  connective  tissue  content.    Thus,  strong 
shadows  were  produced  by  scirrhous  carcinoma 
(rich  in  connective  tissue),  average  shadows  by 
papillary  carcinoma,  and  weak  shadows  by  medullary 
carcinoma  (poor  in  connective  tissue).  The 
comments  of  the  audience  substantiated  the  conclu- 
sion that  increasing  attenuation  was  correlated 
with  increase  in  collagen,  as  in  operative  scars 
and  older  breasts.    Next,  G.  Dale,  Ch.  M.  Gros, 
M.  Gautherie,  and  B.  Gairard  (Senologie-Hospices 
Civils,  Strasbourg,  France)  carefully  reviewed 
diagnostic  image  features,  in  their  paper  on  echo- 
graphic  syndromes  of  breast  cancer.    This  was 
followed  by  a  mul ti -di sci pi ine  approach  to  the 
detection  of  breast  cancer  by  ultrasonic  techniques, 
with  intercomparison  of  signal-processed  ultra- 
sound transmission  data,  ultrasound  pulse-echo 
information,  and  whole  breast  pathology,  by  E.  K. 
Fry,  N.  T.  Sanghvi ,  and  F.  J.  Fry  (Indiana  Univer- 
sity School  of  Medicine  and  Indianapolis  Center  for 
Advanced  Research,  Indianapolis,  Indiana)  and  H.  S. 
Gallager  (University  of  Texas,  Houston,  Texas). 
Experiments  with  an  excised,  formalin-fixed  breast 
revealed  non-uniform  attenuation  particularly  in 
and  near  the  nipple.    The  audience  asked  about  the 
contributions  of  reflections  (which  were  felt  not 
to  be  large),  the  possibility  of  the  existence  of 
bubbles  and  the  effect  of  formalin,  and,  for  in 
vivo  measurements ,  the  effects  of  lactation  and 
the  taking  of  the  contraceptive  pill.    F.  W.  Kremkau, 
C.  P.  McGraw,  and  R.  W.  Barnes  (Bowman  Gray  School 
of  Medicine,  Winston-Salem,  North  Carolina)  reported 
the  results  of  some  careful  measurements  of  the 
acoustic  properties  of  human  brain.    The  infant 
brain  had  about  60  percent  lower  attenuation  than 
the  adult.    In  adult  brain,  fixing  with  formalin 
increased  velocity  by  about  30  percent.  Fixed 
brain  had  a  dispersion  of  about  1.6  m  s'^MHz"^,  and 
fresh  brain  about  2  m  s"^MHz-i.    The  temperature 
coefficient  of  attenuation  was  negative,  and 
attenuation  was  1.4  times  greater  in  white  matter 
than  in  gray.    Tissue  characterization  using 
acoustic  transmission  and  scattering  parameters  was 
further  discussed  in  a  joint  paper  by  M.  P.  Kadaba, 
W.  P.  Cockerill,  and  P.  K.  Bhagat  (University  of 
Kentucky,  Lexington,  Kentucky),  and  R.  W.  Ware 
(Veterans  Administration  Hospital,  Lexington, 
Kentucky).    In  tissues  such  as  kidney,  liver,  and 
cardiac  muscle,  over  the  frequency  range  to  1  to 
10  MHz,  attenuation  increased  from  about  2  to  16  dB 
cm-^,  and  velocity,  from  1520  to  1580  m  s"'  (i.e., 
by  about  4  percent).    In  another  joint  paper,  D.  H. 
Le  Croissette,  R.  C.  Heyser,  P.  M.  Gammell,  and 
J.  A.  Roseboro  (California  Institute  of  Technology, 
Pasadena,  California),  and  R.  L.  Wilson  (University 
of  Southern  California,  Los  Angeles,  California) 
reported  values  of  the  attenuation  of  selected 
tissue  as  a  function  of  frequency.    Using  time-delay 
spectrometry,  involving  the  measurement  of  the 
frequency  difference  between  the  received  signal 


8 


and  a  transmitted  frequency-modulated  wavetrain, 
they  studied  attenuation  over  the  range  1  to  8  MHz 
in  liver,  pancreas,  muscle,  and  fat.    In  liver, 
for  example,  attenuation  decreased  after  death  but 
increased  as  a  result  of  fixing.  Apparently 
similar  tissues  often  had  very  different  attenua- 
tions.   Finally  in  Session  12,  D.  Hughes,  L.  A. 
Geddes,  and  V.  Newhouse  (Purdue  University,  West 
Lafayette,  Indiana)  discussed  the  velocity  and 
attenuation  of  ultrasound  in  blood  at  37  °C.  These 
data  were  required  in  order  to  estimate  Young's 
modulus  of  the  aortic  wall.    Besides  presenting 
accurate  values,  Hughes  reported  that  attenua- 
tion increased  with  packed  cell  volume  (up  to  at 
least  60  percent  packing),  and  velocity  had  a 
minimum  value  with  a  packed  cell  volume  of  about 
10  percent. 

The  final  session.  Session  13,  Tissue  Viability 
and  Tissue  Phantoms,  began  with  a  paper  by  L.  Weiss 
(Roswell  Park  Memorial  Institute,  Buffalo,  New  York) 
on  tissue  signatures  --  a  matter  of  life  and  death. 
Dr.  Weiss  discussed  the  changes  which  took  place  in 
1n  vitro  tissues,  and  the  precautions  which  needed 
to  be  taken  to  minimize  consequential  changes  in 
tissue  signature  interactions.    For  example,  poor 
oxygenation  might  result  in  death  within  5  minutes. 
Even  in  life,  many  tumors  had  dead  or  dying  cells, 
and  these  differences  probably  affect  ultrasonic 
imaging.    Mechanically-induced  cell  separation 
measurements  were  related  to  degenerative  changes 
in  tissue,  and  might  possibly  be  used  to  standardize 
interactions  with  ultrasound.    The  remaining  three 
papers  in  this  Session  were  concerned  with  tissue 
equivalent  materials  for  ultrasonic  imaging.    In  a 
joint  paper,  P.  Edmonds,  Z.  Reye,  and  D.  Parkinson 
(Stanford  Research  Institute,  Menlo  Park,  Califor- 
nia), R.  Filly  (University  of  California  Medical 
Center,  San  Francisco,  California),  and  H.  Busey 
(Picker  Corporation,  Northford,  Connecticut) 
described  a  human  tissue  phantom  for  testing  con- 
ventional ultrasonic  scanners.    After  experimenting 
with  several  rubber  and  plastic  materials,  it  was 
found  that  gelatin-water  gels,  loaded  with 
glass  microspheres  or  cellulose  fillers  as 
scatterers,  gave  satisfactory  results.  Questions 
from  the  audience  were  answered  by  statements  that 
the  gel  had  an  acceptable  dependence  of  attenua- 
tion on  frequency,  and  that  the  plastic  materials 
had  attenuations  which  were  too  high  to  permit 
the  addition  of  scatterers.    R.  C.  Eggleton  (Indiana 
University  School  of  Medicine  and  Indianapolis 
Center  for  Advanced  Research,  Indianapolis,  Indiana) 
mentioned  the  application  of  tissue  simulators  for 
ultrasonic  diagnosis,  in  teaching  and  training, 
evaluation  of  scanner  performance,  and  as  models 
for  basic  research.    He  presented  results  obtained 
with  a  plastisol,  and  showed  the  effects  of  adding 
scatterers.    A  joint  paper  by  P.  L.  Carson,  L. 
Shabason,  and  D.  E.  Dick  (University  of  Colorado 
Medical  Center,  Denver,  Colorado)  and  W.  dayman 
(Alderson  Research  Laboratories,  Inc.,  Denver, 
Colorado)  on  tissue  equivalent  test  objects  for 
comparison  of  ultrasound  transmission  tomography 
by  reconstruction  and  pulse  echo  imaging,  also 
discussed  special  plastic  materials.  Although 
these  materials  did  have  rather  high  attenuation 
(about  2  dB  cm-iMHz"-),  scatterers  could  be  added 
and  phantoms  satisfactory  for  testing  systems 
could  be  constructed.    Finally,  Dr.  Carson  showed 
an  ultrasonic  CT  scan  of  an  in^  vi vo  breast  indica- 
ting an  attenuation  coefficient  of  about  15  dB  cm"^ 


at  3.5  MHz.  • 

In  summary,  the  Second  International  Symposium 
on  Ultrasonic  Tissue  Characterization  brought  the 
subject  into  focus  and  perspective.    The  286  par- 
ticipants, including  many  of  the  leaders  in  the 
field,  spent  three  days  exchanging  ideas  and 
learning  from  each  other.    Mass  screening  for 
breast  cancer  in  particular  was  identified  as  one 
of  several  clinical  problems  which  might  be  solved 
by  ultrasonic  tissue  characterization.  Research 
opportunities  were  evident  in  the  development  of 
fundamental  theories,  the  acquisition  of  data  on 
velocity,  attenuation  and  scattering,  and  in 
clinical  validation.    Image  reconstruction,  signal 
processing,  pattern  recognition,  and  Doppler  blood 
flow  signals,  emerged  as  fruitful  areas  for  inves- 
tigation.   New  technological  advances,  for  example, 
in  tissue  phantom  materials,  were  identified  as 
already  being  ready  for  transfer  to  commercial 
application. 

M.  Linzer 

P.  N.  T.  Wells 

July,  1977 


9 


REPORT  ON  PANEL  DISCUSSION 
ULTRASONIC  DIAGNOSIS  OF  BREAST  CANCER 


PANEL 
Cochairpersons 


William  Pomerance 
Diagnosis  Branch 

Division  of  Cancer  Biology  and  Diagnosis 
National  Institutes  of  Health 
Bethesda,  Maryland 


P.  N.  T.  Wells 

Department  of  Medical  Physics 
Bristol  General  Hospital 
Bristol,  United  Kingdom 


Panel  Members 


Gilbert  Baum 

Albert  Einstein  College  of  Medicine 
Bronx,  New  York 

John  L.  Doppman 

Department  of  Diagnostic  Radiology 
Clinical  Center 
National  Institutes  of  Health 
Bethesda,  Maryland 

G.  H.  Glover 

General  Electric  Company 
Medical  Systems  Division 
Milwaukee,  Wisconsin 

Toshiji  Kobayashi 
Department  of  Internal  Medicine 
National  Cancer  Center  Hospital 
Tokyo,  Japan 

D.  E.  Robinson 
Ultrasonics  Institute 
Sydney,  Australia 


G.  J.  Dale 

Senologie-Hospices  Civils 
Strasbourg,  France 

Elizabeth  K.  Fry 

Indiana  University  School  of  Medicine  and 
Indiana  University  Hospital 
Indianapolis,  Indiana 


J.  F.  Greenleaf 
Mayo  Clinic 
Rochester,  Minnesota 


Melvin  Linzer 

National  Measurement  Laboratory 
National  Bureau  of  Standards 
Washington,  D.C. 

Leonard  Weiss 

Roswell  Park  Memorial  Institute 
Buffalo,  New  York 


In  his  introductory  remarks,  P.  N.  T.  Wells 
pointed  out  that  the  overall  mortality  from  breast 
cancer  was  the  main  cause  of  death  in  Western  women 
between  the  ages  of  40  and  44  years,  and  the  annual 
economic  cost  of  the  disease  in  the  U.S.  was  around 
$200M.    Consequently,  it  was  timely  to  review  the 
contribution  which  ultrasonic  diagnostics  might 
make  to  the  solution  of  the  breast  cancer  problem. 
There  were  many  different  malignant  tumors  of  the 
breast  —  scirrhous,  mammary  duct,  papillary, 
medullary,  colloid,  lobular,  intracystic,  apocrine 
and  adenoid  cystic  carcinoma,  Paget' s  disease, 
lymphoma,  and  sarcoma.    The  earlier  that  breast 
cancer  was  treated,  the  better  was  the  prognosis. 
A  patient  treated  with  in  situ  carcinoma  was  cured. 
If  the  tumor  was  localized,  the  five-year  survival 
was  85  percent;  but  this  fell  to  53  percent  if 
there  was  lymph  node  involvement.    A  patient  with 
distant  metastases  was  incurable,  although  the 
survival  time  varied  from  patient  to  patient. 
Interest  in  screening  hinged  around  the  possibili- 
ty that  earlier  detection  might  improve  prognosis. 
Unfortunately,  thermography  was  unsatisfactory  as 
a  screening  method  for  breast  cancer:    in  one 
series,  for  example,  only  25  percent  of  those  who 
developed  cancer  within  18  months  were  detected, 
and  the  false  positive  rata  was  approaching  15  per- 


cent.   When  combined  with  clinical  examination,  in 
another  trial,  only  about  20  percent  of  cancers  in 
women  under  50  years  of  age  would  not  have  been 
detected  without  mammography.    Moreover,  mammogra- 
phy, whilst  having  a  negligible  risk  as  far  aS 
the  individual  patient  was  concerned,  would  itself 
induce  by  radiation  a  large  number  of  cancers  if 
every  member  of  a  large  population  were  to  be 
exposed  to  it. 

William  Pomerance  said  that  it  had  to  be 
admitted  that  contemporary  methods  of  treating 
advanced  breast  cancer  were  ineffective,  and  that 
progress  might  depend  on  earlier  detection.  He 
defined  screening  as  the  application  of  known 
methods  to  an  asymptomatic  population  to  detect 
early  cancer.    He  considered  that  an  acceptable 
method  for  breast  cancer  screening  would  need  to 
detect  tumors  of  less  than  5  mm  diameter,  and 
that  it  should  be  free  from  hazard  and  have  at 
least  a  90  percent  success  rate.    The  assessment 
of  the  value  of  breast  cancer  screening  would  take 
a  long  time,  since  the  mortality  of  patients  sur- 
viving treatment  only  coincided  with  that  of  the 
normal  population  after  about  17  years.    He  thought 
that  screening  might  have  its  biggest  impact  in 
detecting  slowly  growing  asymptomatic  tumors. 


11 


The  fact  that  screening  for  cancer  of  the 
uterine  cervix  had  resulted  in  a  reduction  in 
mortality  led  Leonard  Weiss  to  hope  that  breast 
cancer  screening  might  also  be  effective.  Moreover, 
he  was  encouraged  by  results  which  he  had  obtained 
using  a  high-resolution  ultrasonic  scanner,  that 
it  might  soon  become  possible  to  detect  lesions 
of  less  than  2  mm  diameter,  although  it  might  be 
difficult  to  distinguish  histologically  between 
different  types  of  lesion.    The  detection  of  2  mm 
diameter  lesions  would  be  an  important  but  numeri- 
cally modest  advance  from  the  present  ability  to 
detect  5  mm  lesions  by  manual  palpation;  an  improve- 
ment of  10  percent  in  effectively  treated  patients 
might  result.    Weiss  stressed  the  importance  of 
growth  rate  in  determining  prognosis.    He  then 
referred  to  a  study  of  the  feasibility  of  breast 
cancer  screening,  which  had  been  carried  out  in 
1971.    If  the  38  million  women  in  the  U.S.  who 
might  have  benefited  from  screening  were  to  have 
been  given  this  advantage,  it  had  been  estimated 
that  the  annual  cost  then  would  have  been  around 
$620M.    He  felt  that  it  would  therefore  be  necessary 
to  restrict  screening  to  women  known  to  be  at  risk, 
or  over  50  years  old. 

The  role  of  mammography  was  discussed  by  J .  L. 
Doppman.    He  showed  some  typical  mammograms,  and 
explained  that  the  method  was  best  for  visualizing 
microcalcif ication.    Three-dimensional  display  was 
needed.    A  great  advantage  of  mammography  was  that 
it  could  be  used  by  the  surgeon  to  ensure  that  he 
removed  the  entire  lesion.    Recent  developments  had 
resulted  in  a  reduction  in  the  x-ray  exposure, 
which  presently  was  generally  less  than  0.01  Gy 
(1  rad). 

Next,  experience  with  pulse-echo  ultrasonic 
methods  was  reviewed.    Toshiji  Kobayashi  had  worked 
for  five  years  in  Japan  on  the  ultrasonic  visuali- 
zation of  the  breast.    For  Japanese  women,  scanning 
through  a  flexible  membrane  at  the  bottom  of  a 
water-bath  was  satisfactory.    Kobayashi  described 
the  distinctive  echographic  features  of  lesions 
of  differing  histologies,  and  he  emphasized  the 
importance  of  retrotumorous  shadowing  in  deducing 
information  about  the  attenuation  by  the  lesion. 
Such  was  the  confidence  which  was  felt  in  the 
method  in  Japan  that  specially  equipped  minibuses 
were  being  used  in  mass  screening  trials.  In 
the  breast  cancer  clinic,  both  pulse-echo  and 
Doppler  investigations  were  complementary  to 
thermography  and  mammography,  according  to  G.  J. 
Dale.    He  discussed  the  ultrasonic  scans  of  two 
typical  patients,  and  concluded  that  in  breast 
cancer  diagnosis  the  false  positive  rate  was  0.6 
percent;  the  false  negative  rate  was  11  percent, 
but  this  fell  to  5  percent  when  combined  with 
clinical  examination.    Using  pulsed  Doppler,  he 
had  traced  breast  arteries  of  greater  than  1  to 
2  mm  in  diameter,  and  he  explained  that  Doppler 
information  related  more  to  function  than  to 
structure.    He  had  not  studied  the  capillary 
system  of  malignant  tumors  by  the  Doppler  method. 
Gilbert  Baum  described  the  interpretation  of 
breast  echograms.    Not  only  was  textural  analysis 
of  breast  tissue  patterns  difficult  because  of 
their  wide  variations,  but  also  the  skin  of 
different  individuals  seemed  often  to  transmit 
ultrasound  differently.    He  felt  that  it  was  neces- 
sary to  scan  through  water,  particularly  because 
lesions  might  otherwise  be  displaced  by  the  pressure 


of  the  ultrasonic  probe.    Coupling  through  a  water 
bath  also  allowed  a  large  aperture  transducer  to  be 
used,  and  the  scanning  motion  could  be  automatic. 
Apart  from  the  need  to  develop  instrumentation 
capable  of  yielding  reproducible  results,  the  main 
problem  was  not  to  obtain  the  scans,  but  to  find 
the  time  to  analyze  them.    Furthermore,  it  was  sub- 
jectively apparent  that  image  degradation  occurred 
when  digitized  either  to  fewer  than  360  x  360 
pixels,  or  to  less  than  32  gray  levels.  Sixty 
scans  were  obtained  for  each  patient,  and  to  reduce 
the  analysis  time  he  had  tried  color-coding,  and 
the  superimposition  of  seven  serial  scans  on  a 
three-dimensional  optical  hologram.    Even  with  such 
techniques,  it  was  obvious  that  screening  could 
only  be  offered  to  selected  groups  of  patients. 
Elizabeth  K.  Fry  had  constructed  a  versatile 
scanner  for  research  into  breast  disease.  This 
was  important,  because  the  modification  of  instru- 
ments primarily  designed  for  the  examination  of 
other  structures  was  unsatisfactory.    The  patient 
lay  prone,  her  breasts  immersed  in  water.  A 
variety  of  transducers,  operating  at  different 
frequencies,  had  been  used  to  make  many  scans, 
producing  a  great  range  of  patterns  for  analytical 
study.    Cautious  optimism  was  expressed  by  D.  E. 
Robinson  because  of  his  experience  with  a  water- 
immersion  breast  scanner  using  a  4  MHz  transducer, 
with  an  aperture  of  40  mm  diameter  and  a  focal 
length  of  100  mm.    Both  simple  and  compound  scans 
had  been  made  with  this  instrument.    The  ultra- 
sonic breast  scan  pattern  depended  on  the  age  of 
the  patient.    Cysts  could  easily  be  demonstrated, 
and  so  could  some  parts  of  the  ductal  system. 
Robinson  mentioned  eight  features,  such  as  the 
degree  of  shadowing,  for  which  scans  of  solid 
lesions  required  to  be  examined  in  reaching  a 
diagnosis.    It  was  easier  to  see  skin  dimpling 
associated  with  malignancy  in  immersed  breasts, 
because  of  the  extra  support  provided  by  the  water. 

The  possible  role  of  ultrasonic  computed 
tomography  in  obtaining  data  on  tissue  properties 
was  emphasized  by  J.  F.  Greenleaf.    In  any  particu- 
lar breast,  it  should  be  possible  to  measure  the 
attenuation  and  velocity  in  the  constituent  image 
pixels,  and  these  data  could  be  used  to  correct 
conventional  two-dimensional  pulse-echo  scans. 
G.  H.  Glover  had  previously  talked  about  his  work 
on  ultrasonic  computed  tomography  in  a  paper 
presented  during  Session  6.    As  time  was  short. 
Glover  did  not  add  to  the  information  which  he  had 
already  given  concerning  the  increase  in  velocity 
associated  with  malignancy  in  breast  tissue.  For 
the  same  reason.  Wells  did  not  elaborate  on  the 
potential  of  neovascularization  blood  flow  Doppler 
detection,  as  he  had  mentioned  this  in  the  paper 
which  he  had  already  presented  in  Session  4. 

Melvin  Linzer  emphasized  that  computed 
reflection  and  transmission  tomography  coupled 
with  ray  tracing  and  frequency-dependent  time-gain- 
compensation  techniques  would  be  needed  to  make 
ultrasonic  breast  screening  feasible.  Because 
of  the  complexity  of  these  calculations  and  the 
vast  quantities  of  data  which  would  be  obtained 
in  scanning  only  one  patient,  ultrafast  processers 
with  large  memory  are  required.    He  felt  that  the 
potential  advantages  of  ultrasonic  investigation, 
and  especially  the  possibility  of  safe  serial 
surveys,  justified  the  expenditure  of  substantial 
resources  on  research. 


12 


Members  of  the  Panel  and  the  audience  then 
began  an  open  discussion.    M.  L.  Skolnik  asked 
whether  it  was  possible  to  demonstrate  microcalci- 
fication  on  ultrasonic  scans,  and  Dale  replied 
that  this  was  so  if  the  calcification  was  suffi- 
ciently extensive.    Fry  pointed  out  that  the  echo- 
graphy was  very  helpful  in  monitoring  young  women 
with  "lumpy"  breasts,  because  a  suggestion  of  the 
development  of  malignancy  could  be  discovered 
early  on;  Robinson  agreed  with  this.  Pomerance 
asked  Fry  whether  she  felt  that  mammography  had 
anything  to  contribute  where  a  lesion  had  been 
detected  by  ultrasound,  and  although  she  thought 
that  this  might  occasionally  be  the  case,  Weiss 
did  not.    C.  R.  Hill  returned  to  the  value  of 
echography  in  managing  patients  with  benign  lesions, 
but  he  felt  that  the  method  was  not  helpful  if  the 
lesions  were  very  small.    Greenleaf  stressed  the 
importance  of  data  obtained  from  time-of-f 1 ight 
computed  tomography.    Fry  considered  that  past 
progress  more  than  justified  continued  efforts  to 
develop  an  ultrasonic  breast  screening  method;  but 
there  was  general  agreement  with  Baum  when  he  said 
that  the  presently  available  commercial  equipment 
ought  not  to  be  used  in  a  trial  of  ultrasonic 
breast  cancer  screening  as  this  would  inevitably 
bring  the  method  into  disrepute.    F.  E.  Barber 
enquired  about  the  potential  of  low-dose  mammog- 
raphy, and  Doppman  replied  that  the  accuracies  of 
such  techniques  had  not  been  assessed;  he  thought 


that  it  would  be  better  to  wait  for  a  viable  ultra- 
sonic method,  than  to  extend  the  use  of  mammography. 
According  to  Pomerance,  even  if  the  x-ray  exposure 
were  to  be  reduced  to  2  mGy,  mammography  would  be 
unacceptable  for  mass  screening.    A  member  of  the 
audience  asked  whether  it  was  the  tumor  which  was 
demonstrated  by  echography,  or  the  reaction  of  the 
surrounding  tissue.    Robinson  replied  that  both 
features  might  be  seen;  for  example,  besides  the 
echoes  from  the  tumor,  there  might  be  skin  thick- 
ening and  shadowing.    In  this  connection.  Fry 
mentioned  the  change  in  the  collagen  content  of 
the  entire  breast,  which  might  be  due  to  a  malig- 
nant tumor.    Wells  closed  the  discussion  with  a 
short  summary. 

He  said  that  it  seemed  to  have  been  estab- 
lished that  mass  screening  of  selected  groups  of 
the  female  population  for  breast  cancer  ~  those 
specially  at  risk,  and  those  more  than  50  years 
old  --  would  be  worthwhile,  if  a  suitable  technique 
could  be  developed.    Such  a  technique  might  well 
be  based  on  ultrasound  but  it  would  not  necessarily 
be  a  simple  pulse-echo  method.    It  would  need  to 
be  accurate  and  economic.    The  present  lines  of 
research  were  promising;  more  resources  should  be 
devoted  to  solving  the  fundamental  and  technologi- 
cal problems,  so  that  earlier  clinical  application 
would  be  possible. 

P.  N.  T.  Wells 
July,  1977 


13 


INTRODUCTORY  ADDRESS 


THE  RELATIONSHIP  OF  TISSUE  SIGNATURE 
RESEARCH  TO  IMPROVED  HEALTH  CARE 

Alfred  J.  Eggers ,  Jr. 

Assistant  Director  for  Research  Applications 
National  Science  Foundation 
Washington,  DC  20550 


It  is  a  pleasure  to  be  here  today  before  such 
a  distinguished  audience.     It  has  been  only  two 
years  since  the  first  conference  devoted  to  ultra- 
sonic tissue  characterization.    Much  has  been 
accomplished  in  those  two  years.    Even  more  will 
be  done  in  the  future.    You  should  be  proud  to  be 
part  of  such  a  vigorous  research  effort,  and  I 
know  you  share  my  pleasure  at  the  prospects  for 
the  improved  medical  devices  that  should  derive 
from  the  efforts  of  all  of  you. 

It  is  because  of  these  prospects  that  the 
applications  side  of  the  National  Science  Founda- 
tion is  so  interested  in  your  efforts.     In  addi- 
tion, there  is  the  revolution  in  microelectronics 
which  has  provided  an  order  of  magnitude  reduction 
in  cost  and  improved  performance  every  five  years 
for  well  over  25  years.    Our  technologists  confi- 
dently expect  another  decade  of  the  same.  What 
does  this  mean?    Suppose  a  highly  sophisticated 
piece  of  equipment  using  the  current  tissue  signa- 
ture technology  would  cost  $1,000,000  on  the 
market  today.     I'm  told  that  this  is  a  reasonable 
ball  park  estimate.    Five  years  from  now  the  cost 
could  plummet  to  $1 00,000--sti 1 1  a  bit  rich  for 
the  average  physician.    But  in  another  five,  we 
find  that  $100,000  may  drop  to  $1 0,000--much 
closer  to  what  can  be  afforded. 

Reflect  for  a  moment  on  what  the  implications 
of  these  changes  could  be.    We  may  be  approaching 
the  end  of  the  "visible"  computer  because  micro- 
electronics can  make  a  computer  so  small  and 
inexpensive  that  it  will  disappear  behind  the 
imaging  screen.    One  question  is  how  will  we 
exploit  this  capability?    This  is  why  efforts  such 
as  this  conference  are  so  important.    It  is  one 
crucial  step  in  preparing  wisely  for  the  future. 

Of  course,  we  must  be  aware  that  extrapola- 
tion is  sometimes  a  hazardous  experience.  There 
is  no  assurance  that  wanting  or  expecting  a  major 
cost--performance  breakthrough  will  make  that 
change  happen.    Still,  we  would  all  be  derelict  in 
discharging  our  responsibilities  if  we  were  not 
ready.    After  all,  the  sums  being  devoted  to 
research  are  not  trivial,  but  they  are  really  a 
low  cost  insurance  payment  if  the  predicted  gains 
from  electronics  are  realized.    The  real  waste 
would  be  to  have  all  this  great  technology  avail- 
able and  not  be  ready  to  apply  it  for  improved 
health  care.    You  might  be  interested  in  the  back- 
ground for  the  Research  Applied  to  National  Needs 
(RANN)  participation  with  the  National  Bureau  of 


Standards  and  the  National  Institutes  of  Health  in 
support  of  this  conference.    Early  in  the  history 
of  the  applications  directorate,  we  set  up  a 
series  of  experiments  to  investigate  ways  of 
accelerating  the  use  of  technology.    One  is  of 
particular  interest  to  you.    The  first  step  was  to 
set  up  a  study  group  in  1973  which  evaluated  the 
state  of  development  of  medical  ultrasonic  diag- 
nostic instruments.    This  survey  team  found  that 
exi sti ng  technology  could  be  marshalled  into 
improved  systems.    Our  group  of  RSD  incentives 
experiments,  therefore,  included  an  Experiment  No. 
5  to  encourage  the  industrial  community  to  create 
these  new  systems.    As  in  most  experiments,  we 
have  learned  and  did  not  do  everything  to  per- 
fection.   Overall,  however,  it  was  a  useful 
investigation.    Today  there  is  new  equipment  on 
the  market  and  clinical  ultrasound  is  rapidly 
becoming  an  important  part  of  diagnostic  services. 

That  same  report  of  1973  also  had  another 
major  finding.    To  make  a  major  step  forward 
beyond  the  experiment  five  objectives  would  take 
far  more  research  across  a  broad  front  of  acti- 
vity.   Later,  we  had  the  Alliance  for  Engineering 
in  Medicine  and  Biology  suggest  in  their  report 
entitled,  "A  Five  Year  Research  and  Development 
Agendum  for  Ultrasonic  Imaging  Diagnostic  Instru- 
mentation," that  a  complete  system  look  across  23 
research  categories  was  needed.     In  broad  terms, 
the  report  pointed  out  the  need  to  balance  the 
total  program  and  consider  it  as  a  system  of 
interrelated  parts.    Funds  for  research  on  bio- 
hazard  were  described  as  needed.    Still,  it  is 
silly  to  not  avail  oneself  of  the  opportunity  to 
acquire  information  at  lower  powers  which  are 
intuitively  less  risky.    Both  types  of  efforts 
were  recommended.     Included  in  this  list  was 
research  on  tissue  characteri zati on--the  subject 
of  this  conference. 

A  direct  outgrowth  of  the  last  tissue  signa- 
ture conference  was  a  recognition  that  ultrasound 
tissue  signature  research  was  not  yet  a  coherent 
body  of  knowledge.    The  RANN  program  provided 
grant  funds  to  establish  and  disseminate  a  pre- 
liminary data  base  for  ultrasound  tissue  charac- 
terization research.     I  understand  that  a  news- 
letter is  being  published  and  that  several  of  you 
in  this  conference  will  meet  on  June  16  to  con- 
tinue the  task  of  systematically  collating  re- 
search results.     I  also  understand  that  the  first 
meeting  will  take  the  place  of  an  advisory  com- 
mittee under  the  leadership  of  Dr.  McKinney,  past 


15 


president  of  the  AIUM  and  professor  of  neurology 
at  the  Bowman  Gray  Medical  School  .    Both  of  these 
groups  are  part  of  a  major  effort  whose  objective 
is  clear.    We  want  improved  patient  care.  Moving 
from  the  present  to  the  future  is  difficult,  and 
we  must  be  sure  that  all  opinions  are  brought  to 
focus  on  this  objective. 

Now,  let  me  point  out  that  we  have  three 
basic  objectives  in  the  RANN  program.    These  are 
to: 

Increase  the  effective  use  of  science  and 
technology  in  dealing  with  national 
probl ems ; 

Shorten  the  lead  time  between  basic 
scientific  discoveries  and  relevant 
practical  applications;  and 

provide  early  warning  of  potential  national 
problems  and  initiative  assessments  and 
research  useful  in  avoiding  or  solving 
such  problems. 

It  is  clear  that  your  research  efforts  sat- 
isfy these  criteria.    You  will  note  that  we  em- 
phasize national  needs.     I  believe  that  each 
country  shares  our  concerns  about  improved  health 
care.     It  is  a  pleasure  to  join  with  all  of  you 
from  many  nations  to  present,  hear,  and  discuss 
common  research  interests  in  this  area.     I  wish 
you  all  the  best  in  your  efforts  today,  tomorrow, 
and  in  the  future  to  use  science  and  technology 
more  effectively  to  make  this  world  a  healthier 
place  in  which  to  live  and  thrive. 


16 


CHAPTER  2 

ATTENUATION  AND  VELOCITY  I:  MECHANISMS 


17 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ELEMENTS  OF  TISSUE  CHARACTERIZATION 


Part  I.    Ultrasonic  Propagation  Properties 


R.  L.  Johnston,  S.  A.  Goss,  V.  Maynard,  J.  K.  Brady, 
L.  A.  Frizzell,  W.  D.  O'Brien,  Jr.,  and  F.  Dunn 

Bioacoustics  Research  Laboratory 

University  of  Illinois 
Urbana,  Illinois    61801,  U.S.A. 


Tissues  can  be  characterized  ultrasonical ly  by  their  attenuation,  absorption,  and 
velocity,  all  of  which  correlate  well  with  the  presence  of  the  major  tissue  com- 
ponents of  water,  and  protein,  particularly,  collagen.    This  correlation  is  examined 
in  solutions  of  biologically  important  molecules  and  in  a  number  of  tissues  and 
organs.    It  is  shown  that  tissues  can  be  grouped  according  to  similar  ultrasonic 
propagation  properties,  physiological  functions,  and  concentration  of  elementary 
constituents.    The  role  of  collagen  in  determining  ultrasonic  properties  of  normal 
and  pathological  tissues  is  discussed. 


Key  words:    Absorption;  amino  acids;  attenuation;  frequency;  mammalian  tissues; 

polypeptides;  proteins;  tissue  characterization;  ultrasonics;  velocity. 


1.  Introduction 

The  ultrasonic  propagation  properties  of  bio- 
logical materials  include  the  behavior  of  those 
measurable  acoustic  parameters,  as  functions  of 
the  state  and  acoustic  variables,  which  charac- 
terize the  fate  of  acoustic  signals  propagating 
within  the  biological  environment.    The  ultra- 
sonic attenuation  includes  not  only  the  absorp- 
tion of  the  ultrasonic  signal,  which  is  degraded 
to  heat,  but  also  losses  due  to  other  mechanisms 
by  which  energy  is  extracted  from  the  propagating 
wave  or  is  redirected  by  virtue  of  the  inhomoge- 
neous  nature  of  the  media.    The  ultrasonic  velo- 
city and  the  characteristic  acoustic  impedance, 
which  can  be  determined  with  the  addition  of 
density  information,  embody  within  them  both  the 
inertial  and  restoring  parameters  of  the  particu- 
lar materials.    Thus,  knowledge  of  the  ultrasonic 
velocity  and  loss  terms  may  provide  a  basis  for 
developing  tissue  signatures  for  various  bio- 
logical materials. 

The  paper  will  deal  with  the  ultrasonic  propa- 
gation properties  of  tissues  beginning  with  the 
elemental  constituents.    As  water,  soft  tissues, 
and  organs  have  very  much  the  same  densities  and 
compressibilities,  it  is  instructive  to  begin  a 
review  with  properties  of  aqueous  media. 

2.    Biological  Molecules  in  Solution 
A.  Water 

The  measured  ultrasonic  absorption  in  water  is 
proportional  to  the  square  of  the  frequency,  over 
the  range  10"^  to  10"^  MHz,  with  the  frequency-free 


absorption  coefficient,  a/f^,  having  a  constant 
value  of  15.7  x  iQ-i^  s2/cm  at  37  °C.    The  magni- 
tude of  this  absorption  is  greater  than  one  would 
expect  from  consideration  of  the  so-called  clas- 
sical absorption  due  to  viscosity  and  thermal 
conductivity.    This  absorption  in  excess  of  the 
classical  value  has  been  attributed  by  Hall  [1]^ 
to  a  structural  relaxation  mechanism  involving  a 
transition  between  two  possible  quasi-crystalline 
states  for  water.    More  recent  experimental  re- 
sults [2]  are  consistent  with  the  hypothesis  that 
water  undergoes  a  structural  relaxation  charac- 
terized by  a  time  constant  of  lO'^^  seconds,  and 
supports  the  idea  that  water  is  a  mixture  of  two 
or  more  states  and  that  the  relaxation  processes 
consist  of  the  independent  jumping  of  molecules 
from  one  state  to  another. 

The  velocity  of  sound  propagation  in  pure  water 
exhibits  a  maximum  at  75  °C  due  to  the  existence 
of  a  minimum  in  the  product  of  density  and  adia- 
batic  compressibility  at  that  temperature  [3]. 
Similar  behavior  is  exhibited  by  dilute  aqueous 
solutions,  although  the  temperature  of  maximum 
velocity  may  be  decreased  since  the  solutes  modify 
the  structural  arrangements  of  water.    The  veloc- 
ity in  water  has  been  measured  most  precisely  by 
McSkimin  [4],  Greenspan  and  Tschiegg  [5],  and 
DelGrosso  and  Mader  [6]. 

B.    Amino  Acids 

Since  it  has  been  observed  that  proteins  play  a 
dominant  role  in  the  absorption  properties  of  tis- 


' Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


19 


sues,  aqueous  solutions  of  amino  acids  and  poly- 
peptides require  attention,  for  completeness. 
When  dissolved  in  water  without  additional  ionic 
constituents  to  influence  the  state  of  charge  of 
the  amino  acid,  the  solution  exhibits  a  magnitude 
of  the  frequency-free  absorption  parameter,  a/f^, 
which  varies  little  with  frequency  in  the  mega- 
hertz range,  and  differs  only  slightly  from  that 
of  the  solvent,  water  [7-12].    Amino  acids  in 
aqueous  solution  at  neutral  pH  may  be  considered 
according  to  their  action  of  structure  making  or 
structure  breaking  in  the  solvent.    Herein,  the 
amino  acids  are  present  as  doubly  charged  molecules 
(zwitterions )  and  are  susceptible  to  dissociation 
and  recombination  reactions  upon  a  change  of  their 
environment.    Hydrogen-bonding  sites  are  located 
on  both  the  amino  and  carboxyl  groups,  while  the 
side  chain  may  be  acidic,  nonpolar,  or  basic. 
Thus,  the  potential  for  breaking  or  making  struc- 
ture, in  the  vicinity  of  the  solute  molecules  are 
considerable.    Hammes  and  Pace  [7]  suggested  that 
the  predominate  ultrasonic  absorption  mechanism  in 
aqueous  solutions  of  glycine,  diglycine,  and  tri- 
glycine  is  that  which  involves  solute-solvent 
(water)  interaction.    When  the  pH  of  an  amino  acid 
aqueous  solution  is  within  the  range  of  2  to  4  or 
11  to  13,  the  relaxational  behavior  can  be  describ- 
ed by  a  single  relaxation  frequency.    A  number  of 
amino  acids  have  been  investigated  as  a  function 
of  pH,  viz. ,  glycine  [9,13-16],  serine  and  threo- 
nine [8],  glutamic  acid,  aspartic  acid  and  alanine 
[14],  and  arginine  and  lysine  [10,15].  Absorption 
maxima  have  been  observed  within  the  pH  ranges 
2  to  4,  and  11  to  13,  with  such  peaks  being  de- 
scribed quantitatively  by  assuming  that  the  proton- 
transfer  reaction  dominates  the  absorption. 

C.  Polypeptides 

When  the  amino  acids  are  formed  into  polypep- 
tide chains,  the  absorption  increases  dramatically, 
and  the  mechanisms  believed  to  be  responsible  for 
the  absorption  in  tissues  may  begin  to  appear.  In 
general,  aqueous  solutions  of  polypeptides  at 
neutral  pH  exhibit  an  ultrasonic  absorption  behav- 
ior greatly  increased  over  those  of  amino  acids  in 
solution  or  of  water,  and  with  a  somewhat  lesser 
than  the  square-of-f requency  dependence.  The 
absorption  in  aqueous  polypeptide  solutions  may 
involve  any  or  all  of  the  following  four  possible 
mechanisms,  vi z. ,  proton  transfer,  helix-coil 
transition,  solvation,  and  relaxation  of  the  shear 
viscosity.  Major  interest  has  tended  to  be  focused 
on  the  proton-transfer  reactions  [17-20]  and 
helix-coil  transitions  [21-24]. 

D.  Proteins 

Continuing  with  increasing  complexity  of  the 
biological  media,  the  polypeptides  may  be  con- 
sidered to  be  arranged  in  a  particular,  -ubiquitous 
way  to  form  the  globular  proteinaceous  state. 
Hemoglobin  and  serum  albumin  solutions  have  re- 
ceived considerable  attention,  partly  because  of 
the  availabilities  of  the  materials.    Special  con- 
sideration is  given  to  other  macromolecul ar  species 
such  as  nucleic  acids  and  polysaccharides. 

When  biological  macromolecules ,  such  as  the 
globular  protein  hemoglobin  are  in  aqueous  solution, 
a  certain  amount  of  the  solvent  becomes  an  inherent 
part  of  the  molecule  since  the  polymer  possesses 
ionic  and  polar  groups  which  associate  with  water 


molecules.    In  addition,  proteins  contain  a  number 
of  nonpolar  side  chains  such  that,  within  the 
vicinity  of  the  macromolecules,  some  water  struc- 
turing occurs.    Thus,  it  is  possible  that  the 
structure  of  liquid  water,  the  hydration  layer, 
increases  in  the  neighborhood  of  the  biological 
macromolecule.    It  is  considered,  therefore,  that 
as  an  acoustic  wave  propagates  through  an  aqueous 
solution  of  biopolymers,  it  perturbs  the  hydration 
layer  manifesting  absorption  of  energy  by  a  struc- 
tural relaxation  process.    The  role  of  molecular 
conformation  of  the  biopolymer  has  also  been  con- 
sidered as  the  origin  of  the  observed  ultrasonic 
absorption.    Figure  1  shows  the  excess  frequency- 
free  absorption  per  unit  concentration  for  several 
biomacromolecules  and  supports  the  view  that  struc- 
turing contributes  to  the  ultrasonic  absorption 
spectra  [25].  Both  dextran  [26],  a  carbohydrate, 
and  polyethylene  glycol  [27],  a  synthetic  polymer, 
assume  random  coil  configurations  in  aqueous  solu- 
tion and  exhibit  absorption  magnitudes  similar  to 
that  of  gelatin.    Hemoglobin  has  a  quaternary 
structure,  bovine  serum  albumin  and  ovalbumin  have 
tertiary  structure,  polyglutamic  acid  [28],  a  syn- 
thetic polyamino  acid,  has  secondary  structure, 
and  the  double  helix,  DNA,  has  a  rigid  rod  con- 
formation [29].    However,  hemoglobin  in  5  molar 
aqueous  guanidine  hydrochloride  solution  exists 
as  a  random  coil  [30]  and  yet  exhibits  ultrasonic 
absorption  spectra  similar  to  that  of  hemoglobin 
in  aqueous  solutions  [31].    The  importance  of 
molecular  spatial  arrangement  is  thus  an  unsettled 
question,  though  the  polypeptide  structure  appears 
to  be  of  considerable  significance. 


1  1  1 

1   1             '        1      1    1    1  M  1. 

^DNA-SS 

10  °c  : 

^BSA- 

OX,IO°C 

/  \  x>0^\ 

/    ^  > 

/  voO^ 

/  vO- 
-  Hb-0X,IO°C 

^  .OM 

-ox,io°c 

f  DNA-CT,I0°C^"^  ^ 

^PGA  pH  5.8  ,25°C  -_ 

■^^                    DEXTRAN, 20°C  - 

-    PEG,  20.7  "C^ 

/Hb,25°C 

PEG,  4.2°C 

,1 

1    1^  t  1  1  1  1  1  1   

ll  I  I  l^l  I  1 1 1 1  1  I   

10  100  1000 

frequency  (  MHz ) 


Fig.  1.    Excess  frequency-free  ultrasonic  absorp- 
tion per  unit  concentration  [25]. 

Within  figure  1,  bovine  serum  albumin  and  hemo- 
globin have  a  molecular  weight  of  68,000  while 
that  of  ovalbumin  is  46,000.    e-lactoglobin  [32] 
(not  on  graph)  with  a  molecular  weight  of  35,000, 
would  fall  in  the  range  determined  by  hemoglobin, 
bovine  serum  albumin,  and  ovalbumin,  while  lyso- 
zyme  [32]  (not  on  graph)  with  a  molecular  weight 
of  14,600,  would  appear  between  hemoglobin  at 
25  °C  and  dextran.    Two  random  coil  polymers, 
dextran  and  polyethylene  glycol,  have  been  studied 
as  a  function  of  molecular  weight.    For  dextran 
solutions,  the  frequency-free  absorption  per  unit 
concentration  increases  with  increasing  molecular 
weight  to  a  molecular  weight  of  about  10,000, 
which  corresponds  to  approximately  100  monomer 
units,  beyond  which  it  is  independent  of  molecular 


20 


weight  [33].    Aqueous  polyethylene  glycol  solu- 
tions show  similar  absorption  behavior  in  that  be- 
yond a  molecular  weight  of  about  4500,  which  also 
corresponds  to  a  chain  length  of  about  100  monomer 
units,  the  absorption  is  independent  of  molecular 
weight  [27].    Thus,  ultrasonic  absorption  depends 
to  some  degree  upon  molecular  weight.  Possibly 
beyond  about  100  monomer  units,  the  macromolecule 
assumes  random  coil  characteristics. 

It  is  uncertain  whether  the  absorption  mechan- 
ism(s)  responsible  for  energy  loss  in  biological 
polymer  solutions  (usually  of  concentrations  less 
than  10  percent  by  weight)  are  the  same  as  those 
responsible  for  the  absorption  properties  of  tis- 
sue [34].    Kremkau  and  Carstensen  [11]  have  sug- 
gested that  more  highly  concentrated  macromolecular 
solutions,  promoting  solute-solute  ( intermolecul ar ) 
interactions,  may  better  approximate  the  tissue 
environment,  and  from  the  point  of  view  of  biologi- 
cal effects,  may  be  the  most  important  level  at 
which  the  ultrasound  interacts.    In  the  case  of 
dilute  solutions,  where  the  absorption  is  con- 
sidered to  vary  linearly  with  polymer  concentra- 
tion, the  mechanism  of  absorption  is  attributed  to 
processes  involving  the  interaction  between  the 
solvent  (usually  water)  and  the  solute.    In  other 
than  dilute  solutions,  the  absorption  is  found  to 
increase  nonlinearly,  and  it  has  been  suggested 
that  this  phenomenon  is  due  to  intermol ecul ar  in- 
teraction processes  [34-36].    It  has  recently  been 
shown  [37],  from  absorption  measurements  of  bovine 
serum  albumin,  to  concentrations  of  40  g/100  ml, 
over  the  frequency  range  3.4  to  15  MHz,  and  at 
20  °C,  that  the  absorption  dependence  upon  con- 
centration does  not  possess  a  linear  region. 
Here,  the  excess  ultrasonic  absorption  can  be  de- 
scribed adequately  by  a  concentration  dependence 
to  the  1.2  power.    It  thus  appears  essential  to 
consider  intermolecular  interaction  over  the  en- 
tire range. 

3.  Tissues 

The  nearly  linear  frequency  dependence  of  at- 
tenuation in  liver,  kidney,  brain,  muscle,  fat, 
and  other  parenchyma,  are  considered  in  the  ap- 
proximate frequency  range  of  100  kHz  to  100  MHz, 
and  the  recent  findings  with  regard  to  the  tem- 
perature dependence  are  discussed.    Several  tis- 
sues and  organs,  such  as  bone,  lung,  refractive 
media  of  the  eye,  and  collagenous  tissues,  are 
singled  out  for  special  detailed  consideration. 


It  is  shown  that  classification  of  tissues  accord- 
ing to  certain  ultrasonic  propagation  properties 
can  be  carried  out  with  regard  to  water  and  col- 
lagen content,  and  with  regard  to  certain  teleo- 
logical  considerations.    Pulmonary  tissue  is  an 
exception  here. 

A.    Frequency  Dependence 

The  dependence  of  the  ultrasonic  absorption  co- 
efficient upon  the  acoustic  frequency  has  been 
studied  by  numerous  investigators.    Goldman  and 
Hueter  [38]  prepared  an  early  compilation  of  ultra- 
sonic velocity  and  absorption  data  in  mammalian 
soft  tissues.    Therein  it  is  seen  that  the  veloci- 
ty, excluding  lung,  is  very  nearly  that  of  dilute 
salt  solutions  and  varies  only  slightly  among 
those  tissues.    Fatty  tissues  are  exceptions  in 
having  a  velocity  about  10  percent  less  than  the 
others.    Figure  2,  taken  from  their  paper,  is  a 
graphical  representation  of  the  acoustic  amplitude 
absorption  coefficient  per  wavelength  for  several 
mammalian  tissues  in  the  frequency  range  of  ap- 
proximately 200  kHz  to  10  MHz.    As  they  attempted 
to  include  all  measurements  available  at  that  time, 
by  numerous  investigators  employing  different  ex- 
perimental techniques,  the  scatter  of  the  data 
exhibited  by  the  bands,  or  broad  shaded  regions  is 
thus  not  wholly  surprising  since  many  neglect  to 
give  either  complete  specifications  of  their  ex- 
perimental procedure  or  a  description  of  the  state 
of  the  specimen  used.    For  example,  it  is  not  pos- 
sible, in  many  cases,  to  determine  from  the  litera- 
ture the  temperatures  employed  by  the  investigators 
reporting  the  data.    It  is,  however,  possible  to 
discern  several  relatively  simple  relationships. 
For  example,  the  absorption  per  cycle,  a/f,  is 
generally  constant  over  the  frequency  range  con- 
sidered.   For  fat,  a/f  increases  slightly  in  the 
frequency  range  from  1  to  10  MHz.    The  experiment- 
al results  for  striated  muscle  and  liver  appear  to 
exhibit  a  minimum  in  the  neighborhood  of  2  MHz. 
Kessler  [39]  has  shown  that  for  kidney  the  linear 
dependence  of  the  attenuation  on  frequency  exists 
to  about  100  MHz,  after  which  a  square  law  (or 
greater)  dependence  exists.    Fry  [40]  has  con- 
sidered a  viscous  mechanism  for  the  absorption  of 
ultrasound  in  tissue,  in  which  it  is  shown  that 
the  viscous  forces  acting  between  a  suitably  chosen 
distribution  of  suspended  particles  (or  structural 
elements)  and  a  suspending  liquid  can  account  for 
the  experimentally  observed  linear  relationship  be- 


21 


tween  acoustic  absorption  coefficient  and  ultra- 
sonic frequency.    The  frequency  band  over  which 
linearity  obtains  (in  the  model)  is    determined  by 
the  limits  of  the  distribution  Of  values  for  the 
parameters  chosen  to  describe  the  structural  ele- 
ments.   Below  the  linear  range  the  theory  predicts 
a  quadratic  dependence,  in  agreement  with  ex- 
periment. 

Figure  3  is  another  way  of  presenting  the  data 
which  may  be  suggestive  for  determination  of 
mechanisms  of  absorption  [41].    Here,  the  loga- 
rithm of  the  absorption  coefficient  is  plotted  as 
a  function  of  the  logarithm  of  the  ultrasonic  fre- 
quency, and  the  slopes  of  the  resulting  curves  are 
examined  (the  slope  is  the  exponent  of  frequency 
upon  which  the  magnitude  of  the  absorption  coeffi- 
cient depends).    Figure  3  shows  data  for  several 
materials  of  increasing  biological  complexity, 
exhibiting  correspondingly  increasing  complexity 
in  absorptive  behavior.    The  urea  solution  exhibits 
a  slope  of  2,  indicative  of  classical  viscous 
absorption  for  which  a/f^  is  a  constant.  Homogen- 
ized milk,  a  suspension  of  fat  particles  and 
hydrated  casein  complexes,  exhibits  a  slope  of 
nearly  unity  from  approximately  1  to  40  MHz.  Be- 
havior of  this  type  cannot  be  explained  in  terms 
of  simple  viscosity  or  scattering  theories.  The 
curves  of  the  absorption  coefficients  of  egg  al- 
bumin, brain  tissue,  liver,  and  striated  muscle 
(not  shown  in  fig.  3)  exhibit  slopes  between  1  and 
2  in  the  neighborhood  of  1  MHz  and  approach  a  slope 
of  2  at  higher  frequencies.    Hueter  [41]  has  sug- 
gested that  this  type  of  frequency  dependence  can 
be  described  for  specific  muscle  preparations  by 
a  double  relaxation  process  in  which  the  bulk 
(volume)  viscosity  of  the  tissue  possesses  a  relax- 
ation frequency  near  40  kHz,  and  the  shear  viscosity 
possesses  a  relaxation  near  400  kHz.    Although  it  is 
conceded  that  this  is  an  oversimplification  of  a 
complicated  process,  it  will  be  shown  below  that  the 
temperature  dependence  of  the  acoustic  absorption 
coefficient  lends  support  to  this  view. 


m  1.0 

-o  8 

■S  6 

I  I 

§  3 
c 

t  2 


0.1 


0.02 


1    1  1  1  1  1 1 1 

whole  liver,  -^ 
1  hour  \ 

fresh  brain  — 

1     1    1  1  1  y  1 

//  /  / 
/  / 
/  //  / 
1  //^XT\\\Y.  / 

/      '//  ' 
/  / 

/  /  - 
/ ^ ^  _ 

liver  homog 

1  hour  / 

/  // — y- — 

/A  / 
//  / 

/''^  ^  egg  white  / 



lOm.urea^  / 

1      /  1  1  1  1  1  1 

1      1  1 

0.1 


1.0 


10 


*50 


frequency  in  (MHz) 


Fig.  3. 


Acoustic  amplitude  absorption  coefficient 
versus  frequency  for  materials  of  differ- 
ent biological  complexity  [41]. 


B.    Temperature  Dependence 

Details  of  the  absorption  coefficient  as  a 
function  of  temperature  and  frequency  have  re- 
cently become  available.    Figure  4  shows  observa- 
tions on  mammalian  central  nervous  system,  the 
only  tissue  for  which  such  data  are  available. 


0.8 


O.G 


e 

o 


0.2 


0.26  MHz 


0.5  MHz 
0.7  MHz 

"^'^^  4.2  MHz 


Fig.  4.    Frequency  and  temperature  dependence  of 

ultrasonic  absorption  in  mammalian  central 
nervous  tissue  [43]. 

The  curves  for  0.26,  0.5,  0.7,  and  1  MHz  rep- 
resent in  vivo  measurements  in  the  spinal  cords 
of  neonatal  mice  (essential  poi ki 1 otherms ) 
[42-44]  and  those  for  4.2  MHz  are  in  vitro 
measurements  in  brains  of  adult  cats  (homeo- 
therms)  [45].    The  relatively  complex  behavior  of 
the  frequency-free  absorption  coefficient  with 
frequency  and  temperature  suggests  a  family  of 
curves  whose  maxima  decrease  in  magnitude,  and  oc- 
cur at  ever  higher  temperatures,  as  frequency  in- 
creases, supporting  the  suggestion  mentioned 
above.    It  is  not  known  whether  other  soft  tis- 
sues exhibit  similar  behavior  but  Kishimoto  [46] 
has  observed  a  positive  temperature  coefficient 
for  the  absorption  coefficient  of  bone  in  the 
frequency  range  1.4  to  4.5  MHz.    These  data  il- 
lustrate the  necessity  for  complete  specification 
of  the  state  of  specimens  when  reporting  experi- 
mental results. 

C.    Absorption  and  Velocity  in  Bone 

Bone  is  a  tissue  possessing  acoustic  propaga- 
tion properties  greatly  different  from  those  of 
the  soft  tissues  discussed  previously.    An  early 
study  of  specially  prepared  skull  bone,  in  the 
frequency  range  0.6  to  3.5  MHz  (25  to  35  °C), 
yielded  a  quadratic  dependence  of  the  absorption 
coefficient  upon  frequency  with  a  transition  to  a 
linear  dependence  beyond  about  2  MHz  [47].  An 
average  value  found  for  the  acoustic  amplitude 


22 


absorption  coefficient  per  unit  path  length  in 
skull  bone,  in  the  neighborhood  of  1  MHz,  was  of 
the  order  of  1  cm"^,  approximately  an  order  of 
magnitude  greater  than  that  of  soft  tissues  of  the 
same  temperature.    However,  recent  observations 
have  called  these  values  into  question,  and  Adler 
and  Cook  [48]  have  obtained  absorption  measure- 
ments of  1.5  cm"^  and  2.2  cm"^  in  freshly  frozen 
dog  tibia  at  room  temperature  at,  respectively, 
3  and  5  MHz.    Reports  of  measurements  of  the  longi- 
tudinal speed  of  sound  in  bone  are  largely  in 
agreement  that  it  is  approximately  twice  that  of 
soft  tissues  [46,48-50].    Anisotropy  of  elastic 
properties  and  variations  in  density  of  bone 
present  special  problems  for  measurement  and  for 
interpretation  of  results  [48-50]. 

D.    Refractive  Media  of  the  Eye 

Begui  [51]  has  studied  the  acoustic  properties 
of  the  refractive  media  of  the  eye  in  vitro.  He 
determined  the  ultrasonic  absorption  coefficient 
of  the  aqueous  and  vitreous  humors  at  30  MHz  and 
that  of  the  lens  at  3  MHz.    The  specimens  were  ob- 
tained from  excised  fresh  calf  eyes.    At  30  MHz 
and  27.5  °C,  the  aqueous  and  vitreous  humors  both 
exhibit  an  acoustic  amplitude  absorption  coeffi- 
cient of  0.35  cm"^.    Since  this  is  approximately 
50  percent  greater  than  the  absorption  coefficient 
of  dilute  salt  solutions,  it  suggests  that  the  ab- 
sorption coefficients  of  the  humoral  media  of  the 
calf  eye  possess  a  viscous-type  dependence  upon 
frequency;  that  is,  the  absorption  coefficient 
probably  increases  as  the  square  of  the  frequency. 
The  lens  of  the  calf  eye  exhibits  a  value  of  0.7 
cm"-'  for  the  acoustic  amplitude  absorption  coef- 
ficient at  3  MHz  and  28  °C.    Since  the  lens  con- 
tains a  relatively  high  concentration  of  protein, 
it  is  reasonable  to  assume,  in  the  absence  of 
further  information,  that  the  frequency  dependence 
of  the  absorption  coefficient  of  the  lens  re- 
sembles that  of  other  soft  tissue  for  which  the 
absorption  appears  to  be  dominated  by  the  protein 
content,  i.e.,  it  is  probable  that  the  absorption 
coefficient  per  unit  path  length  of  the  lens 
varies  approximately  with  the  first  power  of  the 
frequency.    Some  investigators  currently  using 
ultrasonic  methods  for  diagnosing  disorders  of 
the  human  eye  feel  that  the  lens  absorption  value 
given  by  Begui  is  larger  than  that  for  the  human 
lens  in  vivo.    The  possible  discrepancy  may  result 
because  of  species  differences.    Indeed,  Begui 
observed  that  the  viscosity  of  the  intraocular 
fluid  of  calf  eyes  is  greater  than  the  values 
normally  stated  for  the  fluid  media  of  human  eyes. 
Further,  the  specimens  used  by  Begui  were  first 
stored  (at  temperatures  in  the  neighborhood  of 
0  to  5  °C)  and  were  used  for  measurement  purposes 
within  a  time  interval  of  10  days.    Begui  obtained 
for  the  speed  of  sound  in  refractive  media  of  the 
eye  1497  m/s  for  the  aqueous  humor,  1516  for  the 
vitreous  humor,  and  1616  for  the  lens. 

E.    Pulmonary  Tissue 

Two  earlier  studies  [52,53]  showed  that  ultra- 
sonic attenuation  in  freshly  excised  dog  lung  was 
unusually  high,  that  the  speed  of  sound  was  con- 
siderably less  than  that  of  water,  and  that  both 
of  these  quantities  had  a  strong  dependence  on 
pulmonary  inflation  and  acoustic  frequency.  It 
was  also  shown  that  a  pathological  condition  in- 


volving an  accumulation  of  liquid-like  matter 
within  the  pulmonary  architecture,  had  the  effect 
of  appreciably  reducing  both  the  attenuation  and 
the  velocity.    Two  recent  studies  have  provided 
details  of  the  frequency  and  inflation  dependencies. 
Dunn  [54]  has  shown  that  for  excised  dog  lung  of 
inflation  to  a  fraction  of  residual  air,  such  that 
the  specimen  density  is  0.4  g/cm^,    the  attenuation 
increases  exponentially  from  4  cm"^  at  1  MHz  to 
12  cm~i  at  5  MHz.    In  this  same  frequency  range, 
the  speed  of  sound  increased  linearly  from  0.66  x 
10^  cm/s  to  1.2  X  10^  cm/s.    These  findings  are  in 
general  agreement  with  those  of  Bauld  and  Schwan 
[55]  who  also  showed  that,  for  fixed  inflated 
specimens,  the  energy  reflected  at  the  lung-liquid 
interface  ever  increases  the  gaseous  inflation 
allowing  for  lesser  amounts  of  energy  to  enter 
the  lung. 

4.    Role  of  Collagen 

Collagen  is  the  most  abundant  single  protein  in 
the  human  body  and  the  most  common  protein  in  the 
animal  kingdom.    It  is  closely  associated  in  con- 
nective tissue  of  vertebrates  and  comprises  be- 
tween one-quarter  and  one-third  of  the  total 
protein  in  the  human  body,  being  about  six  per- 
cent of  the  total  body  weight  [56].    However,  more 
than  the  prevalence  of  collage  in  the  body,  there 
is  some  evidence  to  suggest  that  its  contribution 
to  the  elastic  properties  of  most  soft  tissues, 
together  with  other  structural  proteins,  deter- 
mines acoustic  contrast  during  echographic  visuali- 
zation [57,58].    This  hypothesis  is  based  on  the 
fact  that  the  static  or  low- frequency  elastic 
modulus  of  collagenous  fibers  is  at  least  1000 
times  greater  than  those  of  soft  tissues.  Since 
the  ultrasonic  velocity  is  proportional  to  the 
square  root  of  the  elastic  modulus,  collagenous 
tissues  are  thought  to  introduce  a  greater  im- 
pedance mismatch  than  would  be  the  case  for  a  soft 
tissue  interface,  thereby  increasing  the  acoustic 
reflectivity.    The  increased  deposition  of  collagen 
and  the  concomitant  increase  in  attenuation  seen 
in  many  pathological  conditions  is  a  basis  for 
ultrasonic  differential  diagnosis. 

Table  1  contains  ultrasonic  attenutati on ,  ve- 
locity, water  content,  total  protein  content,  and 
collagen  content  for  various  tissues  [59].    It  is 
apparent  that  the  greater  the  collagen  content, 
the  greater  the  attenuation.    Dependence  of  at- 
tenuation and  velocity  upon  water  content  are  al- 
so apparent.    These  data  allow  empirical  relations 
to  be  formed  for  more  quantitative  assessment  of 
the  role  of  collagen  content  of  tissues  upon  their 
ultrasonic  propagation  properties.    For  example, 
figure  5  represents  a  summary  of  table  I  of  the 
ultrasonic  attenuation  at  1  MHz  as  a  function  of 
the  wet  weight  percentage  of  collagen  for  ten  tis- 
sues.   Using  linear  regression  by  the  method  of 
least  squares,  a  reasonable  fit  to  the  data  is 
provided  by  the  relation 

A  =  0.17  C°-3,  (1) 

where  A  is  the  ultrasonic  attenuation  in  cm"^  and 
C  is  the  wet  weight  percentage  of  collagen.  The 
best  fit  parameter,  the  coefficient  of  determina- 
tion, r^,  yields  a  value  of  0.71  (unity  represents 
a  perfect  fit).    Equation  1  is  represented  on 
figure  5  by  the  solid  straight  line.  Logarithmic, 
exponential,  and  linear  functions  were  also  analyz- 


23 


Table  1.    Ultrasonic  attenuation  and  velocity  for  tissues  of  various 
water,  protein  and  collagen  content. 

Tissue  Attenuation  at      Velocity  Water  Protein  Collagen 

1  MHz  (cm-i)  (m/s)  (%)  (%)  (%) 


Water  (20  °C) 

0.0003 

1483 

100 

Amniotic  fluid 

0.0008 

1510 

97 

0.27 

Agueous  humor 

0 

10  -  0.017 

QQ 

u . UU J   -  1 

Vitreous  humor 

0 

10  -  0.017 

1  D  1  0 

QQ 

yy  - 

QQ 

yy 

y 

U  .       -  U .  £13 

0.014  - 

0.06; 

r  c  r 
Lor 

0.0012 

1  A  QQ 

1  0  1  D 

QQ 

yy 

(J,  UJ 

PI  a sma 

0.01 

1  0/  1 

y  u  - 

QR 

y  D 

7 

Tes  ti s 

0.019 

(absorption) 

Q/l 

trace 

Blood 

0.02 

1571 

74  - 

83 

Milk 

0.04 

1485 

87 

3  -  4 

Fat 

0 

04  -  0.09 

1410  - 

1479 

10  - 

19 

5  -  7 

yes 

Spl een 

0.06 

1520  - 

1591 

76  - 

80 

17  -  18 

0.5 

-1.2 

Liver 

0 

07  0.13 

1  550  - 

1607 

68  - 

78 

20  -  21 

0.1 

-1.3 

Kidney 

0 

09  -  0.13 

1558  - 

1568 

76-83 

15  -  17 

0.5 

-  1.5 

Brain 

0 

09  -  0.13 

1510  - 

1565 

75  - 

79 

10  10 

0.04 

-  0.3 

Spinal  cord 

0 

09  -  0.12 

64  - 

80 

Striated  muscle 
against  grain 
with  grain 

0 
0 

18  -  0.25 
08  -  0.12 
0.16 

1  DDO  - 

1592  - 
1  576  - 

1  cnQ 
1  oUi 

1603 

1587 

66  - 

80 

on  91 

0.7 

-  1.2 

Heart 

0 

25  -  0.38 

1572 

77  - 

78 

17 

0.4 

-  1.6 

Tongue 

against  grain 
with  grain 

0.58 
0.28 

1  ETC 
1  0/0 

1585 

62  - 

68 

14  -  17 

Lens 

0 

10  -  0.20 

1616 

63  - 

69 

30  -  36 

Articular  capsule 

0.38 

Integument 

0.40 

1498 

60  - 

72 

7 

-  30 

Cartilage 

0.58 

1665 

23  - 

34 

70 

49  -  63 

10 

-  20 

Tendon 

against  grain 
with  grain 

0.54 
0.58 

1750 

63 

35 

32 

0.01 


tissue  collagen  (percentage) 


Fig.  5.    Attenuation  at  1  MHz  as  a  function  of  the  percentage  of  tissue 
collagen  for  10  tissue  types  [59]. 


24 


ed  but  yielded  worse  fits  than  equation  1.  Simi- 
larly, to  a  first  approximation,  the  ultrasonic 
velocity  was  examined  as  a  function  of  collagen 
content  for  8  tissues,  excluding  integument,  and 
yielded  the  expression 

C  =  -1700  +  230  In  v,  (2) 

where  v  is  the  ultrasonic  velocity  in  meters/second 
and  r^  =  0.91.  Again,  first  approximations  for  wet 
weight  percentage  of  total  protein,  P,  yielded 

A  =  0.004  pi-26,  r2  =  0.77,  (3) 

for  16  biological  materials  excluding  integument. 
Thus  to  a  first  approximation  there  appear  to  be 
mathematical  relationships  which  can  be  developed 
to  relate  the  amount  of  tissue  constituents  to  the 
ultrasonic  propagation  properties.    There  are  some 
tissues,  such  as  fat  and  integument,  which  may 
have  to  be  treated  separately  but,  otherwise,  this 
approach  suggests  that  such  relationships  may  aid 
in  developing  ultrasonic  tissue  signatures  which 
can  be  incorporated  into  clinical  instrumentation. 

5.    Concluding  Remarks 

What  emerges  from  all  this  is  summarized  in 
table  2,  following  Dussik  and  Dunn  [50,61],  which 
is  an  attempt  to  characterize  tissues  according 


to  their  ultrasonic  propagation  properties  and 
biological  function.    It  is  seen  that  tissues  can 
be  grouped  according  to  apparent  teleology  fashion 
with  relatively  narrow  ranges  of  attenuation 
values  within  each  group.    The  attenuation  approxi- 
mately doubles  from    group  to  group  in  the  direc- 
tion of  increasing  attenuation,  and  the  speed  of 
sound  increases  in  the  same  direction.  Further, 
proceeding  from  group  to  group  in  the  same  direc- 
tion, tissues  of  ever-decreasing  water  content  and 
ever-increasing  structural  protein  content  become 
included.    Thus,  it  is  seen  that  ultrasonic  at- 
tenuation and  velocity  may  be  invoked  to  charac- 
terize tissues  according  to  functional,  structural, 
and  teleological  criteria.    Possibly  detailed  mea- 
surements will  allow  assignment  of  resolvably 
unique  values  to  each  tissue  structure,  including 
usefully  differentiable  values  for  pathological 
states.    Should  this  be  the  case,  ultrasonic 
attenuation  and  impedance  values,  as  functions 
of  state  and  acoustic  parameters,  media,  etc . , 
should  specify  uniquely  tissues  for  diagnostic 
purposes . 

Acknowledgment 

The  authors  acknowledge  gratefully  the  partial 
support  for  portions  of  the  activities  described 
herein  by  grants  from  the  National  Institutes  of 
Health. 


Table  2.    Average  attenuation  of  tissues  by  categories. 


Tissue 
attenuation 
categories 


Attenuation 
at  1  MHz 
(cm-i) 


Ti  ssue 


Assumed 
teleology 


General 
trends 


1.  Very  low 

2.  Low 

3.  Medium 

4.  High 

5.  Very  high 


0.03 
0.01 

0.06-0.07 


0.08-0.11 
0.11 

0.08-0.16 

0.23 

0.3 

0.4 
0.5 
0.6 

1  or  more 
>  4 


serum 
blood 

adipose  tissue 


nervous  tissue 
1  i  ver 
muscle 
heart 
kidney 

integument 
tendon 
carti lage 

bone  (mineralized) 
pulmonary  tissue 


ion  ,  metabol ic ,  etc . , 
transport  convection 

energy  and  (water) 
storage 

physiological 
function 
parenchymal 
tissue 


structural 
integration 
stromal  tissues 

skeletal  framework 

gaseous  exchange 


Increas-  Increas- 
ing ing 
struc-  speed 
tural  of 
protein  sound 
content 


Increas- 
ing 
H2O 

content 


References 

[1]    Hall,  L.,  The  origin  of  ultrasonic  absorp- 
tion in  water,  Phys.  Rev.  73,  775  (1948). 

[2]    Davis,  C.  and  Jarzynsky,  J.,  Liquid  Water- 
Acoustic  Properties:    Absorption  and  Re- 
laxation, Chapter  12  in  The  Physics  and 
Physical  Chemistry  of  Water,  Vol .  I  of  Water: 
A  Comprehensive  Treatise,  F.  Franks,  ed., 
(Plenum  Press,  New  York,  1972). 

[3]    Willard,  G.  W. ,  Temperature  coefficient  of 


ultrasonic  velocity  in  solutions,  J.  Acoust. 
Soc.  Am.  19,  235  (1947). 

[4]    McSkimin,  H.  J.,  Velocity  of  sound  in  distil- 
led water  for  the  temperature  range  20  °C  - 
75  °C,  J.  Acoust.  Soc.  Am.  37,  325  (1965). 

[5]    Greenspan,  M.  and  Tschiegg,  C.  E.,  Speed  of 
sound  in  water  by  a  direct  method,  J.  Res. 
Natl.  Bur.  Std.  59  249-254  (1957). 

[6]    DelGrosso,  V.  A.  and  Mader,  C.  W. ,  Speed  of 
sound  in  pure  water,  J.  Acoust.  Soc.  Am.  52 , 


25 


1442-1446  (1972). 

[7]    Hammes,  G.  G.  and  Pace,  C.  N.,  Ultrasonic 
absorption  measurements  in  aqueous  solutions 
of  glycine,  diglycine,  and  triglycine, 
J.  Phys.  Chem.  72,  2227  (1968). 

[8]    White,  R.  D.,  Slutsky,  L.  J.,  and  Pattison, 
S.,  Kinetics  of  the  proton  transfer  reac- 
tions of  serine  and  theronine,  J.  Phys. 
Chem.  25,  161  (1971). 

[9]    Hussey,  M.  and  Edmonds,  P.  D.,  Ultrasonic 
examination  of  proton-transfer  reactions  in 
aqueous  solutions  of  glycine,  J.  Acoust. 
Soc.  Am.  49,  1309  (1971). 

[10]    Hussey,  M.  and  Edmonds,  P.  D.,  Ultrasonic 
examination  of  proton-transfer  reactions  at 
the  a-amino  and  side-chain  groups  of  arginine 
and  lysine  in  aqueous  solution,  J.  Acoust. 
Soc.  Am.  49,  1907  (1971). 

[11]    Kremkau,  F.  W.  and  Carstensen,  E.  L. , 

Macromolecular  Interaction  in  Sound  Absorp- 
tion, in  Interaction  of  Ultrasound  and 
Biological  Tissues--Workshop  Proceedings, 
J.  M.  Reid  and  M.  R.  Sikov,  eds.,  DHEW 
Publication  (FDA)  73-8008  BRH/DBE  73-1, 
p.  37  (U.S.  Government  Printing  Office, 
Washington,  D.C.  1972). 

[12]    Kremkau,  F.  W.,  Macromolecular  Interaction 
in  the  Absorption  of  Ultrasound  in  Biologi- 
cal Material,  Ph.D.  Thesis,  University  of 
Rochester  (1972). 

[13]    Maass,  G.,  Ph.D.  Thesis,  University  of 
Gottingen  (1962). 

[14]    Applegate,  K. ,  Slutsky,  L.  J.,  and  Parker, 

R.  C,  Kinetics  of  proton-transfer  reactions 
of  amino  acids  and  simple  polypeptides, 
J.  Am.  Chem.  Soc.  90,  6909  (1968). 

[15]    Hussey,  M. ,  Proton  Transfer  at  Side  Chain 
Groups:    A  Mechanism  for  the  Absorption  of 
Ultrasound  in  Protein  Solutions,  Ph.D. 
Thesis,  University  of  Pennsylvania  (1970). 

[16]    Hussey,  M. ,  Proton  Transfer  Processes:  One 
of  the  Modes  of  Interaction  of  Ultrasound 
with  Proteins  and  DNA,  in  Interaction  of 
Ultrasound  and  Biological  Tissues,  J.  M. 
Reid  and  M.  R.  Sikov,  eds.,  DHEW  Publication 
(FDA),  73-8008  BRH/DBE  73-1,  p.  31  (U.S. 
Government  Printing  Office,  Washington,  D.C. 
1972). 

[17]    Zana,  R.  and  Tondre,  C,  Ultrasonic  studies 
of  proton  transfers  in  solutions  of  poly- 
(lysine)  and  poly(ornithine) .  Implications 
for  the  kinetics  of  the  helix-cord  transi- 
tion of  polypeptides  and  for  the  ultrasonic 
absorption  of  proteins,  J.  Phys.  Chem.  76 
1737  (1972). 

[18]    Lang,  J.,  Tondre,  C,  and  Zana,  R. ,  Effect 
of  urea  and  other  organic  substances  on  the 
ultrasonic  absorption  of  protein  solutions, 
J.  Phys.  Chem.  75,  374  (1971). 


[19]  Zana,  R.  and  Lang,  J.,  Effect  of  pH  on  the 
ultrasonic  absorption  of  aqueous  solutions 
of  proteins,  J.  Phys.  Chem.  74,  2734  (1970). 

[20]    Applegate,  K. ,  Slutsky,  L.  J.,  and  Parker, 
R.  C,  Kinetics  of  proton-transfer  reactions 
of  amino  acids  and  simple  polypeptides,  J^. 
Am.  Chem.  Soc.  90,  6909  (1968). 

[21]    Doty,  P.,  Wada,  A.,  Yang,  J.  T.,  and  Blout, 
E.  R.  ,  Polypeptides:    VIII,  molecular  con- 
figurations of  poly-L-glutamic  acid  in 
water-dioxane  solution,  J.  Polymer  Sci.  23, 
851  (1957). 

[22]    Barksdale,  A.  D.  and  Stuehr,  J.  E.,  The 
kinetics  of  the  helix-coil  transition  in 
aqueous  poly-L-glutamic  acid,  J.  Am.  Chem. 
Soc.  94,  3334  (1972). 

[23]    Schwarz,  G.,  On  the  kinetics  of  helix-coil 
transition  of  polypeptides  in  solution, 
J.  Mol.  Biol.  n_,  64  (1965). 

[24]    Wada,  Y.,  Sasabe,  H.,  and  Tomono,  M. , 

Viscoelastic  relaxations  in  solutions  of 
poly-(glutamic  acid)  and  gelatin  at  ultra- 
sonic frequencies,  Biopolymers  5,  887  (1967). 

[25]    Dunn,  F.,  Acoustic  Properties  of  Biological 
Materials,  in  Fundamental  and  Applied 
Aspects  of  Nonionizing  Radiation,  S.  M. 
Michaelson  and  M.  W.  Miller,  eds.,  p.  21 
(Plenum  Publishing  Corporation,  New  York, 
1975). 

[26]    Hawley,  A.  S.,  Kessler,  L.  W.,  and  Dunn,  F.  , 
Ultrasonic  absorption  in  aqueous  solutions 
of  high-molecular-weight  polysaccharides, 
J.  Acoust.  Soc.  Am.  38,  521  (1965). 

[27]    Kessler,  L.  W.,  O'Brien,  W.  D.,  Jr.,  and 
Dunn,  F.,  Ultrasonic  absorption  in  aqueous 
solutions  of  polyethylene  glycol,  J.  Phys. 
Chem.  74,  4096  (1970). 


[28]    Burke,  J.  J.,  Hammes,  G.  G.,  and  Lewis, 

T.  B.,  Ultrasonic  attenuation  measurements 
in  poly-L-glutamic  acid  solutions,  J.  Phys. 
Chem.  42,  3520  (1965). 


[29]    O'Brien,  W.  D. ,  Christman,  C.  L.,  and  Dunn, 
F.,  Ultrasonic  investigation  of  aqueous 
solutions  of  deoxyribose  nucleic  acid, 
J.  Acoust.  Soc.  Am.  52,  1251-1255  (1972). 

[30]    Kawahara,  K. ,  Kirscher,  A.  G.,  and  Tanford, 
C,  Dissociation  of  human  co-hemoglobin  by 
urea,  guanidine  hydrochloride  and  other 
reagent.  Biochemistry  4,  1203  (1965). 

[31]    O'Brien,  W.  D.,  Jr.  and  Dunn,  F. ,  Ultrasonic 
examination  of  the  hemoglobin  dissociation 
process  in  aqueous  solutions  of  guanidine 
hydrochloride,  J.  Acoust.  Soc.  Am.  50,  1213- 
1215  (1971). 

[32]  Lang,  J.,  Tondre,  G.,  and  Zana,  R.,  Effects 
of  urea  and  other  organic  substances  on  the 
ultrasonic  absorption  of  protein  solutions, 
J.  Phys.  Chem.  75,  374  (1971). 


26 


[33]    Hawley,  S.  A.  and  Dunn,  F.,  Ultrasonic  ab- 
sorption in  aqueous  solutions  of  dextran, 
J.  Chem.  Phys.  50^,  3523  (1969). 

[34]    O'Brien,  W.  D. ,  Jr.  and  Dunn,  F.,  Ultra- 
sonic Absorption  by  Biomacromolecules , 
in  Interaction  of  Ultrasound  and  Biological 
Tissues--Worl<shop  Proceedings,  J.  M.  Re  id 
and  M.  R.  Sikov,  eds.,  p.  13,  DHEW  Publica- 
tion (FDA)  73-8008  BRH/DBE  73-1  (U.S.  Govern- 
ment Printing  Office,  Washington,  D.C.  1972). 

[35]    Kessler,  L.  W.  and  Dunn,  F.,  Ultrasonic 

investigation  of  the  conformal  changes  of 
bovine  serum  albumin  in  aqueous  solution, 
J.  Phys.  Chem.  73,  4256  (1969). 

[36]    O'Brien,  W.  D.,  Jr.  and  Dunn,  F.,  Ultrasonic 
absorption  mechanisms  in  aqueous  solutions 
of  bovine  hemoglobin,  J.  Phys.  Chem.  76 , 
528  (1972). 

[37]    Goss,  S.  A.  and  Dunn,  F.,  Concentration  De- 
pendence of  Ultrasonic  Absorption  in  Aqueous 
Solutions  of  Bovine  Serum  Albumin,  in  IEEE 
Ultrasonics  Symposium  Proceedings,  p.  65 
(1974). 

[38]    Goldman,  D.  E.  and  Hueter,  T.  F.,  Tablar 

data  of  the  velocity  and  absorption  of  high- 
frequency  sound  in  mammalian  tissues,  J_. 
Acoust.  Soc.  Am.  28,  35  (1956);  29,  655 
(1957). 

[39]    Kessler,  L.  W.,  VHF  ultrasonic  attenuation 
in  mammalian  tissue,  J.  Acoust.  Soc.  Am.  53 
1759  (1973). 

[40]    Fry,  W.  J.,  Mechanism  of  acoustic  absorption 
in  tissue,  J.  Acoust.  Soc.  Am.  24,  412-415 
(1952). 

[41]    Hueter,  T.  F.,  Viscoelastic  Losses  in  Tissues 
in  the  Ultrasonic  Range,  WADC  Technical  Re- 
port 57-706  (1958). 

[42]    Dunn,  F.,  Temperature  and  amplitude  depend- 
ence of  acoustic  absorption  in  tissue,  J_. 
Acoust.  Soc.  Am.  34,  1545  (1962). 

[43]    Dunn,  F.  and  Brady,  J.  K.,  Pogloshchenie 
ul ' trazvyeka  v  biologicheski kh  sredakh, 
Biofizika  18,  1063  (1973).  Translation, 
Ultrasonic  absorption  in  biological  materials. 
Biophysics  18,  1128  (1974). 

[44]    Dunn,  F.  and  Brady,  J.  K. ,  Temperature  and 
Frequency  Dependence  of  Ultrasonic  Absorp- 
tion in  Tissue,  in  Proceedings  8th  Interna- 
tional Congress  on  Acoustics,  Vol.  I,  p. 
366c  (Goldcrest  Press,  Trowbridge,  Wilts., 
1974). 

[45]    Robinson,  T.  C.  and  Lele,  P.  P.,  An  analysis 
of  lesion  development  in  the  brain  and  in 
plastics  high  intensity  focused  ultrasound 
at  low  megahertz  frequencies,  J.  Acoust. 
Soc.  Am.  51,  133  (1972). 

[46]    Kishimoto,  T. ,  Ultrasonic  absorption  in 
bones,  Acustica  8,  179  (1958). 


[47]    Hueter,  T.  F.,  Messung  der  Ultraschal labsorp- 
tion  in  menschlichen  Schadel knochen  und  ihre 
Abhangigkeit  von  der  Frequenz,  Naturwissen- 
schaften  39,  21  (1952).   

[48]    Adler,  L.  and  Cook,  K.  V.,  Ultrasonic  para- 
meters of  freshly  frozen  dog  tibia,  J. 
Acoust.  Soc.  Am.  58,  1107-1108  (1975T. 

[49]    Thiesmann,  H.  and  Pfander,  F. ,  Uber  die 

Durchlassigkeit  des  Knochens  fiir  Ultraschall, 
Strahlentherapie  80.  607  (1949). 

[50]    Lang,  S.  B.,  Ultrasonic  Method  for  Measuring 
Elastic  Coefficients  of  Bone  and  Results  on 
Fresh  and  Dried  Bovine  Bones,  IEEE  Trans. 
Biomed.  Eng.  BME-17,  101  (1970T^ 

[51]    Begui,  Z.  E.,  Acoustic  properties  of  the  re- 
fractive media  of  the  eye,  J.  Acoust.  Soc. 
Am^  26 ,  365  (1954).   

[52]    Dunn,  F.  and  Fry,  W.  J.,  Ultrasonic  absorp- 
tion and  reflection  by  lung  tissue,  Phys. 
Med.  Biol.  5,  401  (1961). 

[53]    Reid,  J.  M.,  Ultrasonic  Diagnostic  Methods 
in  Cardiology,  Ph.D.  Thesis,  University 
of  Pennsylvania,  1965. 

[54]  Dunn,  F.,  Attenuation  and  speed  of  sound  in 
lung,  J.  Acoust.  Soc.  Am.  56,  1638  (1974). 

[55]  Bauld,  T.  J.  and  Schwan,  H.  P.,  Attenuation 
and  reflection  of  ultrasound  in  canine  lung 
tissue,  J.  Acoust.  Soc.  Am.  56^,  1630  (1974). 

[56]    White,  A.,  Handler,  P.,  and  Smith  E.  L., 

Principles  of  Biochemistry  (McGraw  Hill  Book 
Co.,  New  York,  1968). 

[57]    Fields,  S.  and  Dunn,  F.,  Correlation  of 

echographic  vi sual izabi 1 i ty  of  tissue  with 
biological  composition  and  physiological 
state,  J.  Acoust.  Soc.  Am.  54,  809-812  (1973). 

[58]    Kessler,  L.  W.,  Fields,  S.  I.,  and  Dunn,  F., 
Acoustic  microscopy  of  mammalian  kidney, 
J.  Clinical  Ultrasound  2,  317-320  (1974). 

[59]    O'Brien,  W.  D. ,  The  Role  of  Collagen  in 

Determining  Ultrasonic  Propagation  Proper- 
ties in  Tissue,  in  Acoustical  Holography, 
L.  W.  Kessler,  ed..  Vol.  7  (Plenum  Press, 
New  York,  1977). 

[60]    Dussik,  K.  T. ,  Kyriazidov,  M. ,  Fritch,  D.  J., 
and  Srear,  R.  S.,  Measurement  of  articular 
tissues  with  ultrasound,  Amer.  J.  Phys.  Med. 
37,  160-165  (1958). 

[61]    Dunn,  F.,  Ultrasonic  Attenuation,  Absorption, 
and  Velocity  in  Tissues  and  Organs,  in  Ul tra- 
sonic  Tissue  Characterization,  M.  Linzer, 
ed. ,  National  Bureau  of  Standards  Special 
Publication  453,  pp.  21-28  (U.S.  Government 
Printing  Office,  Washington,  D.C.  1976). 


27 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ABSORPTION  OF  SOUND  IN  TISSUES 


Edwin  L.  Carstensen 

Department  of  Electrical  Engineering 
University  of  Rochester 
Rochester,  New  York    14627,  U.S.A. 


In  spite  of  extensive  applications  of  ultrasound  in  diagnosis,  therapy  and  even 
surgery,  there  are  still  many  problems  to  be  solved  in  the  basic  physics  of  sound 
propagation  in  tissues.    Absorption  of  longitudinal  ultrasonic  waves  occurs  primarily 
at  the  macromolecular  level.    There  is  evidence  to  indicate  that  this  absorption  can 
be  profoundly  modified  by  macromolecular  interaction.    The  specific  structural  or 
chemical  relaxation  mechanisms  responsible  for  the  absorption  are  unknown.  Microscopic 
inhomogeneities  may  lead  to  certain  forms  of  relative  motion  viscous  losses  or  thermal 
absorption.    Macroscopic  inhomogeneities  in  tissue  affect  sound  propagation  and  can 
lead  to  artifacts  in  certain  methods  of  measurement  of  tissue  absorption.    Shear  waves 
are  not  important  in  the  soft  tissues  of  the  body. 

Key  words:    Absorption  of  ultrasound;  macromolecular  relaxation;  relaxation  phenomenon; 
ultrasonic  tissue  absorption. 


Although  acoustic  absorption  phenomena  play 
an  important  role  in  every  medical  application 
of  ultrasound,  we  still  have  only  a  rudimentary 
understanding  of  the  responsible  physical  mech- 
anisms forty  years  after  the  first  work  in  this 
field  [1]^.    This  review  of  basic  concepts  at- 
tempts to  outline  what  we  know  about  these 
processes  and  to  identify  problems  to  be  solved. 

1.    Absorption  Data 

Information  on  absorption  of  sound  in  tissues, 
as  it  is  available  to  us  today,  is  outlined  in 
figure  1.    (A  more  complete  summary  of  absorp- 
tion data  is  included  in  the  paper  by  Johnston 
et  al .  [2].)    Data  for  packed  red  cells  are  in- 
cluded iDecause  we  can  think  of  this  preparation 
as  a  kind  of  simple  model  tissue  and  because  it 
has  been  studied  over  a  wide  range  of  frequencies. 
The  absorption  of  sound  in  fatty  tissue  is 
significantly  lower  than  the  other  soft  tissues. 
Bone  is  much  lossier  than  other  tissues  of  the 
body.    Water  makes  a  negligible  contribution  to 
the  absorption  of  tissues  at  low  frequencies. 
Only  a  few  measurements  have  been  carried  out 
above  10  MHz.    We  can  guess  that  values  of  (aA) 
for  soft  tissues  will  follow  the  pattern  set  by 
hemoglobin  [3]  and  approach  but  remain  signifi- 
cantly above  that  of  water  at  frequencies  of 
the  order  of  1000  MHz.    From  acoustic  microscopy 
it  is  clear  that  sharp  differences  in  absorption 
among  components  of  cells  exist  at  frequencies 
above  1000  MHz  [4]. 

The  range  of  values  which  have  been  reported 
for  a  given  tissue  type  is  large.    This  results 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


in  part  from  normal  biological  variability.  In 
addition,  there  is  a  potential  for  error  in 
measurement  with  i nhomogeneous  tissue  samples. 


10" 


2      5  2      5          2      5         2  5 

0.1  1.0  10  100  1000 

FREQUENCY  (MHz) 


Fig.  1.    Absorption  of  sound  in  biological  mater- 
ials.   Solid  lines  are  estimates  of  mean 
absorption  coefficients  from  many  investi- 
gators.   The  range  of  reported  values  is 
large  particularly  for  bone.    The  high 
frequency  extrapolation  (dashed  line)  for 
concentrated  red  cells  is  based  on  meas- 
urements of  hemoglobin  solutions  [3].  Ex 
Exprapolation  of  the  data  for  soft  tissues 
(dashed  line)  to  100  MHz  is  based  on  meas- 
urements of  kidney  at  100  and  200  MHz 
[35]. 


29 


For  example,  we  found  that  in  measuring  sample 
attenuation  with  a  phase  sensitive  receiver  the 
apparent  absorption  for  muscle  could  be  as  much 
as  four  times  its  true  value  [5].    It  is  inter- 
esting that  the  radiation  force  method  used  by 
Pohlman  [1]  in  the  first  tissue  absorption 
studies  is  still  perhaps  the  most  reliable  tech- 
nique available.    Acoustoel ectric  receivers 
show  great  promise  for  this  kind  of  investiga- 
tion [6] . 

Almost  all  of  the  absorption  studies  have 
used  excised  tissue  samples.    Very  few  in  vivo 
tissue  absorption  measurements  have  been  made; 
but  those  few  are  in  rough  agreement  with 
values  for  excised  tissue.    It  was  reported  that 
freshly  excised  liver  had  a  much  higher  absorp- 
tion than  aged  tissue  samples  at  frequencies 
near  1  MHz  [7,8].    We  have  recently  attempted 
to  duplicate  these  observations  without  success 
[9].    Other  aging  studies  have  shown  very  little 
change  in  absorption  of  tissues  with  time  after 
excision  [10].    From  the  information  available, 
it  appears  reasonable  to  assume  that  the  data 
in  figure  1  are  valid  for  living  tissues,  but 
we  should  realize  that  this  is  an  assumption 
which  has  not  been  thoroughly  tested. 

It  is  a  surprising  admission  that  after  forty 
years  of  study  we  do  not  know  the  temperature 
dependence  of  the  absorption  coefficient  for  all 
of  the  tissues  of  the  body.    The  absorption  co- 
efficients for  blood  [11],  brain  [12],  and  myo- 
cardial muscle  [10]  decrease  slightly  with  tem- 
perature in  the  range  0  to  40  °C.    One  study 
reports  a  positive  temperature  coefficient  for 
bone  between  0  to  60  °C  [13],    As  discussed 
below,  mouse  spinal  cord  is  reported  to  have  a 
complex  temperature  dependence  [14]. 

This  discussion,  of  course,  concerns  longitu- 
dinal ultrasonic  waves  in  tissue.    The  propaga- 
tion constants  for  shear  waves  in  tissues  are  not 
known  in  detail.    However,  recent  rough  measure- 
ments indicate  that  transverse  waves  will  not  be 
generated  in  any  of  the  soft  tissues  of  the  body 
with  any  current  medical  application  of  ultra- 
sound [9,15]. 

2.    Macromol ecul ar  Absorption 

The  primary  absorption  of  ultrasound  in  tissue 
appears  to  take  place  at  the  macromol ecular  level. 
This  was  demonstrated  directly  by  Pauly  and  Schwan 
through  a  comparison  in  the  absorption  of  sound 
in  liver  tissue  and  a  subcellular  homogenate  of 
the  tissue  [16].    The  near  identity  of  the  two 
values  suggests  that  tissue  structure  had  little 
influence  on  the  absorption  process. 

A  simpler  biological  system  yields  more  de- 
tailed information.    The  absorption  coefficients 
for  packed  red  cells  and  for  a  solution  of  hemo- 
globin at  the  same  concentration  in  which  it  is 
found  in  the  erythrocyte  {'^  30  g/100  ml)  are  es- 
sentially identical.    This  is  shown  as  the  upper 
curves  in  figure  2.    Now,  if  a  26  percent  suspen- 
sion of  red  cells  in  physiological  saline  is  pre- 
pared, the  volumes  of  cells  are  the  same  as  in 
the  packed  preparation.    Thus,  the  local  concen- 
tration and  environment  of  the  hemoglobin  in  the 
cell  remains  the  same,  and  hence  the  contribution 
to  the  absorption  from  the  hemoglobin  should  be 
in  direct  proportion  to  amount  of  protein  present 
as  shown  by  the  dashed  line  in  the  figure  2  [17]. 
Saline  contributes  little  to  the  absorption 

30 


2i 


FREQUENCY  (MHz) 


Fig.  2.    Absorption  of  sound  in  a  suspension  of 

bovine  erythrocyte  in  physiological  saline 
[17,18].    Dashed  curve  is  the  macromolecu- 
lar  contribution  to  the  absorption  of  the 
suspension  (26%  cells). 

especially  at  low  frequencies.    Actual  absorp- 
tion measurements  on  such  a  suspension  show  that 
the  contributions  of  the  hemoglobin  solution  in 
the  cell  accounts  for  most,  but  not  all,  of  the 
observed  loss.    The  excess  absorption  which 
arises  from  the  i nhomogeneous  property  of  the 
medium  will  be  discussed  later. 

Since  none  of  the  smaller  molecules  or  ions  in 
the  red  cell  makes  a  significant  contribution  to 
the  absorption,  it  appears  that  the  primary  loss 
comes  from  the  presence  of  the  protein  itself. 
Even  the  amino  acid  building  blocks  of  hemo- 
globin have  much  smaller  specific  absorptions 
(i.e.,  absorption  per  gram  solute)  than  the  in- 
tact protein  as  shown  in  figure  3.    In  fact,  the 
absorption  of  a  10  percent  solution  of  amino 
acids  in  the  proportions  found  in  hemoglobin  is 
barely  distinguishable  from  water  itself,  where- 


1 0  —\  1  1  1  1  1  1  1  1  1 

2        5  2        5  2  5 

105  107  108  109 

f  (Hz) 

Fig.  3.    Absorption  of  sound  in  a  10  percent 

hemoglobin  solution  at  25  °C  [18].  Treat- 
ment with  the  enzyme  pronase  fragmented 
the  protein  yielding  components  with 
molecular  weight  less  than  10,000  and  re- 
sulting in  a  decreased  absorption.  A  prep- 
aration of  amino  acids  in  concentrations 
equivalent  to  that  in  the  hemoglobin  solu- 
tion give  absorption  values  which  are 
hardly  different  than  water  alone. 


as  the  absorption  of  a  10  percent  solution  of 
hemoglobin  is  ten  times  that  of  water  at  10  MHz 
[18].    A  pronase-treated  hemoglobin  solution 
yielding  subunits  with  molecular  weight  less  than 
10,000  showed  a  somewhat  smaller  specific  absorp- 
tion than  that  of  the  intact  protein.    A  more 
subtle  dissection  of  hemoglobin  using  quanidine 
hydrochloride  produced  specific  absorptions  which 
were  both  larger  and  smaller  than  that  of  intact 
hemoglobin  as  the  fractionation  proceeded  [19]. 

The  "whole  much  greater  than  the  sum  of  its 
parts"  is  not  limited  to  hemoglobin  or,  for  that 
matter,  to  proteins.    Figure  4  shows  a  very  simi- 
lar relationship  among  polysaccharides.  The 
absorption  for  dextran  is  much  greater  than  for 
simple  sugars.    It  is  interesting  that  the  spe- 
cific absorption  of  dextran  and  Ficoll  molecules 
of  the  same  molecular  weight  are  roughly  equal. 
Dextran  is  a  long  chain  polymer  whereas  Ficoll 
is  globular.    As  a  result,  the  macroscopic  vis- 
cosity of  equivalent  concentrations  of  the  two 
molecules  is  dramatically  different.    This  is 
just  one  of  many  observations  that  virtually 
eliminate  classical  viscous  processes  from  con- 
sideration in  tissue  absorption. 


^  10-1 

CM 
4- 
U 

o 

^  2 


9  MHz 


sucrose 


5  2 
103 


5  2 
10"* 


105 


Molecular  weight 


Fig.  4. 


Specific  absorption  of  polysaccharides 
vs_.  molecular  weight  [18].    Data  for 
dextran  are  taken  from  reference  [36]. 
Another  polysaccharide,  Ficoll,  which 
is  nearly  spherical  and  thus  has  a  much 
lower  intrinsic  viscosity  than  dextran, 
has  approximately  the  same  specific 
absorption  as  dextran. 


A  number  of  proteins,  nucleic  acids  and  poly- 
saccharides have  now  been  studied  [2].    The  spe- 
cific absorption  values  for  the  macromolecules 
are  uniformly  greater  than  those  associated  with 
the  small  molecules  and  ions  which  make  up  tissue. 

3.    Macromolecular  Interaction 

The  increase  in  specific  absorption  with  mole- 
cular complexity  does  not  appear  to  stop  at  the 
macromolecular  level.    Association  among  macro- 
molecules  in  some  cases  results  in  increased 
values  for  the  absorption  per  molecule.    This  is 
a  vague  loosely  defined  concept  and  is  not  based 
on  any  specific,  postulated  mechanism.  However, 
there  are  a  number  of  observations  which  seem  to 
fit  this  generic  heading.    The  specific  absorp- 
tion of  hemoglobin  increases  dramatically  with 


1Q2- 


5- 


"  "^insulin  Ficoll 
o-dextran  [36] 


2- 


lOi- 


5- 


10°- 


•  - hemoglobin  solution 
-  erythrocyte  suspension  i. 


30  MHz 


10 


— I — 

20 


30 


40 


50 


60 


70 


g  Hb/100  cm3 


Fig.  5. 


Absorption  of  hemoglobin  as  a  function  of 
concentration  at  25  °C  [18].    For  frequen- 
cies greater  than  10  MHz  the  absorption 
coefficients  of  suspensions  of  red  cells 
are  determined  almost  entirely  by  the 
hemoglobin  which  they  contain  [11]. 


concentration  as  shown  in  figure  5  [18].    At  a 
concentration  of  60  grams  hemoglobin  per  100  ml 
solution,  where  the  mean  spacing  of  molecules  is 
very  nearly  equal  to  the  molecular  diameter,  the 
absorption  per  molecule  is  five  times  greater 
than  its  value  in  a  dilute  solution.  Whether 
by  introducing  a  new  mechanism  or  modifying  an 
existing  process,  the  interaction  between  macro- 
molecules  as  the  concentration  increases  causes 
a  marked  increase  in  the  specific  absorption. 

There  are  many  other  clear  examples  where 
macromolecular  interaction  is  associated  with  in- 
creased absorption.    The  hemoglobin  in  rat 
erythrocytes,  when  stored  in  sodium  citrate  at 
4  °C  for  several  days,  becomes  paracrystal 1 ine 
(fig.  6)  [20].    The  specific  absorption  of  para- 
crystalline  cells  is  at  least  twice  that  of 
normal  rat  erythrocytes  [ 18] .    Treatment  of  red 
cells  with  acrolein  cross  links  the  hemoglobin 
and  leads  to  specific  absorption  values  which 
are  four  times  greater  than  the  values  for  normal 
erythrocytes  as  shown  in  figure  7  [21].  Other 
examples  may  be  found  in  figure  8.    It  is  inter- 
esting that  the  sediment  from  homogenized  liver 
has  a  specific  absorption  three  times  greater 
than  the  "supernatant"  from  that  preparation 
[16].    Presumably  the  supernatant  is  a  dilute 
solution  of  macromolecules;  whereas  the  sedi- 
ment consists  of  highly  organized  subcellular 
particles . 

Perhaps  these  observations  are  no  more  than  a 
series  of  coincidences.    However,  the  informa- 
tion available  today  suggests  that  the  trend  to- 
ward increasing  absorption  with  increasing  inter- 
action extends  beyond  the  macromolecular  level 
to  interactions  among  macromolecules. 


31 


acrolein-fixed 
erythrocytes 
■  hypertonic  L 
erythrocytes  J 


•  paracrystalllne  ^ 
rat  erythrocytes  J- 
erythrocyte  '1 
ghost  membranej 

yeast  


normal  rat 
"erythrocytes 

dilute  albumin 

normal  bovine"L 
erythrocytes  J 


-dilute  hemoglobin 


ficoll   

pronase  treated  Hb 


amino  acids 


5.0 


muscle  tissue 
liver  nuclei 


3.0 


liver  tissue 
liver  sediment 


2,0 


1-0 


0.5 


liver  homogenate 


DNA,  RNA 


liver  'supernatant' 


dextran 

(MW  >  30.000) 


0.3 


0.2 


gelatin 


amino  acids 


Fig.  6.    Electron  micrographic  comparison  of  normal 
and  paracrystal 1 ine  rat  erythrocytes. 


103- 


asp 


5- 


rdB/cm] 
Lg/cm^J 


2- 

102- 


IQi 


102 


f  (MHz) 


Fig.  7. 


Specific  absorption  for  normal  (o)  and 
fixed  {o,x,/\)  bovine  erythrocytes  in  aque- 
ous suspension  at  25  °C.    Acrolein  concen- 
trations in  the  original  fixing  solutions 
were  0.3,  1.0  and  2  g/100  cm^.    Cells  were 
washed  free  of  excess  acrolein  before  meas- 
urement [21]. 


Fig.  8.    Comparison  of  specific  absorptions  of 
biological  materials  using  dilute  hemo- 
globin as  a  reference.    All  data  are  for 
10  MHz  at  25  °C  [18]. 


4.    Relaxation  Absorption 

When  stress-strain  relationships  for  a  medium 
depend  upon  shifts  in  chemical  or  structural  equi- 
libria, it  is  possible  that  a  sound  wave  propagated 
in  the  medium  will  experience  relaxation  absorption. 
Specifically,  the  absorption  per  wavelength  is 
[22,23]. 


1  + 


(1) 


where  t  is  the  relaxation  time  of  the  internal  sys- 
tem and  c,  cs,  Cco  are  the  phase  velocity,  and  the 
low  and  high  frequency  limits  of  the  velocity  re- 
spectively.   As  implied  by  the  equation,  the  magni- 
tudes of  the  absorption  and  the  dispersion  in  the 
velocity  are  related.    The  relation  has  been  tested 
quantitatively  for  hemoglobin  [11].    The  results 
are  consistent  with  the  hypothesis  that  a  relaxa- 
tion mechanism  is  responsible  for  the  absorption 
[24].    The  first  velocity  dispersion  measurements 
reported  for  tissues  support  the  hypothesis  that 
relaxation  absorption  is  involved  in  these  mater- 
ials as  well  as  in  macromol ecular  solutions. ^ 


2To  a  very  rough  approximation  we  can  write  [11] 
aA  *  (f/c)(Ac/Af ).    Kremkau  [12]  reports  for 
brain  values  of  aA  *  0.015,  in  agreement  with 
figure  1,  and  velocity  dispersion  of  around 
1  m/s/MHz  near  1  MHz. 


32 


Frequency  (MHz) 

Fig.  9.    Relaxation  absorption.    Log  (aX)  vs^  log 
frequency  is  plotted  for  three  temperatues 
under  assumptions  discussed  in  the  text. 
In  this  example,  the  activation  energy  is 
20  kcal/mol.    The  value  for  (aX)niax  fot" 
the  illustration  is  arbitrary.    With  this 
model,  the  (aX)  curves  simply  shift  to  the 
right,  with  log  coq  proportional  to  the  in- 
crement in  temperature. 

The  frequency  dependence  of  (aX)  given  by  eq. 
(1)  is  shown  in  figure  9.    For  certain  simple  re- 
actions which  may  be  perturbed  by  the  sound  wave, 
we  may  write  [25] 

^  =      .  A  e-E/f^T      ■  (2) 

where  E  is  the  activation  energy  of  the  dominant 
rate  process  in  the  internal  system,  T  the  abso- 
lute temperature,  A  and  R  are  constants.    If  a 
reference  temperature  Tq  is  chosen  so  that 
T  =  To  +  AT,  we  can  write 

log  wo  *  constant  +  -—^       •  (3) 


The  general  behavior  predicted  by  figure  9  is  seen 
in  a  variety  of  simple  systems  i.e.  the  curves 
of  log  (aX)  vs .  log  f  shift  to  the  right  in 
increments  proportional  to  the  change  in  temper- 
ature.   This  is  illustrated  with  MnSO^  solutions 
in  figure  10  [26].    The  temperature  coefficient 
of  the  absorption  is  negative  if  observations 
are  carried  out  somewhat  below  the  relaxation 
frequencies,  while  it  is  positive  at  high  fre- 
quencies.   Figure  11  shows  the  temperature  de- 
pendence of  (aX)  for  relaxation  absorption  with 
activation  energies  of  20  and  50  cal/mol.    As  an 
example,  data  for  lecithin  vesicles  at  1  MHz  are 
shown  [27]. 

Few  biological  materials  can  be  characterized 
by  a  single  relaxation  process.    Rather,  in  most 
cases  there  appear  to  be  broad  distributions  of 
relaxation  frequencies,  in  the  case  of  hemo- 
globin, for  example,  extending  from  below  0.1 
MHz  to  beyond  100  MHz  (fig.  1).    From  our  first 
order  model,  we  would  anticipate  that  increasing 
temperature  would  simply  shift  the  curve  of  log 
(aX)  vs_.  log  f  to  higher  frequencies.    Where  the 
slope  of  this  curve  is  slightly  positive  this 
leads  to  small  negative  temperature  coefficients 


0.0 


0.21  I       I  \  \  I  \  I 

0.2        0.4    0.6     1.0        2.0        4.0   6.0  10.0 

Frequency  (MHz) 

Fig.  10.    Absorption  of  sound  in  0.5  molar  MnSOi, 
solutions  in  water.    Dashed  curves 
represent  the  contribution  of  the  low 
frequency  relaxation  to  the  absorption 
[26]. 

10-2i 


Temperature  (°C) 

Fig.  11.    Relaxation  absorption.    Log  (aX)  vs_. 

temperature  is  plotted  for  activation 
energies  of  20  and  50  kcal/mol.    As  an 
example,  data  (o)  for  lecithin  vessicles 
[27]  are  shown. 


for  the  absorption.    Brain  [12],  blood  [11],  and 
heart  muscle  [10]  (fig.  12)  appear  to  have  this 
characteristic.    The  absorption  per  wavelength 
in  bone,  as  reported  by  Kishimoto  [13]  decreases 
with  frequency  (fig.  13).    As  anticipated,  this 
leads  to  a  positive  temperature  coefficient  for 
the  absorption.    In  all  of  these  materials,  the 
activation  energies  are  of  the  order  of  a  few 
thousand  kilocalories  per  mole. 


33 


The  1n  v1vo  data  for  mouse  spinal  cord  [14] 
are  unique  among  the  limited  sample  of  tissues 
for  which  temperature  dependence  of  absorption 
has  been  studied  (fig.  14).    In  contrast  with 
other  tissues  which  appear  to  have  many  relaxa- 
tion frequencies,  mouse  spinal  cord  at  2  °C 
shows  a  frequency  dependence  of  (aA)  which  is 
too  strong  to  be  explained  even  by  a  single  re- 
laxation frequency.    In  addition,  the  shape  of 
the  log  (aX)  vs.  log  f  curve  changes  with  tempera- 
ture.   Because  of  these  unusual  properties,  mouse 
spinal  cord  appears  to  be  a  particular  candidate 
for  further  study.    Since  temperature  studies 
may  provide  general  information  about  the  proper- 
ties of  internal  relaxing  systems  and  in  a  sense 
extend  the  range  of  frequencies  observed,  it 
would  be  desirable  to  have  this  information  for 
a  broad  range  of  tissues. 


1.0 


Fig.  12. 


2  5 
Frequency  (MHz) 


10.0 


Absorption  of  sound  by  myocardium  [10]. 
The  data  illustrate  the  pairing  of  a 
positive  slope  in  log  (aX)  vs^.  log  fre- 
quency with  a  negative  temperature  coef- 
ficient for  the  absorption. 


10-1 


2- 


10- 


^s ingle 
^v^relaxation 
time 


1.0- 


0.14  ^  ,  1 

12  5  10 

Frequency  (MHz) 

Fig.  13.    Absorption  of  sound  by  bone  [13].  In 
this  case  the  slope  in  log  (aX)  vs^.  log 
frequency  is  negative  and  the  teniperature 
coefficient  of  the  absorption  is  positive 


Frequency  (MHz) 

Fig.  14.    Absorption  of  sound  by  mouse  spinal  cord 
[14].    It  appears  the  observations  at 
2  °C  cannot  be  explained  by  relaxation 
mechanisms. 

Identification  of  the  specific  chemical  or 
structural  reactions  which  are  responsible  for 
relaxation  absorption  in  tissues  remains  the 
pristine  challenge  of  this  field  of  study.  Al- 
though some  effort  has  been  given  to  this  prob- 
lem, we  must  admit  almost  complete  ignorance  of 
the  nature  of  tissue  relaxation  mechanisms  at  a 
meaningful,  basic  level. 

5.    Microscopically  Inhomogeneous  Materials 

Although  the  absorption  of  sound  in  blood  oc- 
curs primarily  at  the  macromolecular  level,  a 
significant,  measurable,  contribution  at  fre- 
quencies below  10  MHz  arises  from  the  presence  of 
intact  red  cells  in  suspension  (fig.  2).  Scat- 
tering, viscous  relative  motion,  and  thermal 
absorption  have  been  considered  as  possible  mech- 
anisms for  this  excess  loss  [17,28-34].  Classi- 
cal scattering  is  ruled  out  on  qualitative  and 
quantitative  grounds.    Relative  motion  absorption 
occurs  when,  because  of  density  differences  be- 
tween suspended  particle  and  suspending  medium. 


34 


there  is  relative  motion  between  the  two  phases 
with  a  consequent  viscous  loss.    A  comparison  of 
observed,  excess,  non-protein  absorption  for  sus- 
pensions of  red  cells  in  physiological  saline 
with  that  predicted  by  theory  for  relative  mo- 
tion absorption  is  shown  in  figure  15  [17]. 


0.4  0.7 


.0        2  4 
Frequency  (MHzI 


20 


Fig.  15.    Comparison  of  non-hemoglobin  absorption 
in  suspensions  of  red  cells  in  saline 
with  that  predicted  by  relative  motion 
absorption  (solid  curves).    Points  are 
taken  from  the  data  in  figure  2  and 
reference  [17]. 

Thermal  absorption  arises  when  thermal  expansion 
coefficients  and  specific  heats  of  particles  and 
the  suspending  medium  differ  significantly  thus 
leading  to  irreversible  heat  flow  between  the  two 
phases.    If  extreme  assumptions  are  made  for  the 
thermal  properties  of  blood  cells,  the  predicted 
thermal  asborption  is  almost  as  large  as  that 
predicted  for  viscous  relative  motion  [33]. 
Definitive  experiments  to  compare  contributions 
from  the  two  mechanisms  have  not  been  attempted. 
Present  evidence,  however,  points  to  relative 
motion  as  the  dominant  process.    The  suggestion 
that  relative  motion  may  be  important  in  the 
solid  tissues  [31,34]  remains  to  be  tested. 

Acknowledgments 

The  author  wishes  to  acknowledge  the  contribu- 
tion of  Dr.  Robert  Weed  for  the  use  of  the  elec- 
tron micrograph  of  a  paracrystal 1 i ne  red  cell. 

This  review  and  much  of  the  research  reported 
has  been  supported  in  part  by  U.S.P.H.S.  Grant 
GM09933. 


References 

[1]      Pohlman,  R.,  Ueber  die  Absorption  des 
Ultraschalls  im  menschlichen  Gewebe  und 
ihre  Abhaengi gkei t  von  der  Frequenz, 
Physik.  Z.  40,  159-161  (1939). 

[2]     Johnston,  R.  L.,  Goss,  S.  A.,  Maynard ,  V., 
Brady,  J.  K. ,  Frizzell,  L.  A.,  O'Brien, 
W.  D.,  Jr.,  and  Dunn,  F.,  Elements  of  Tissue 
Characterization.    Part  I.    Ultrasonic  Propa- 
gation Properties  (this  publication,  p.  19). 


[3]      Edmonds,  P.  D.,  Bauld,  T.  J.,  Ill,  Dyro, 
J.  F.  ,  and  Hussey,  M.,  Ultrasonic  absorp- 
tion of  hemoglobin  solutions,  Biochim.  et 
Biophys.  Acta  200,  174-177  (1969y 

[4]      Lemons,  R.  A.  and  Quate,  C.  F.,  Acoustic 
microscopy:    biomedical  applications. 
Science  188,  905-911  (1975). 

[5]     Marcus,  P.  W.  and  Carstensen,  E.  L., 

Problems  with  absorption  measurements  of 

inhomogeneous  solids,  J.  Acoust.  Soc .  Am. 

58,  1134-1335  (1975). 

[6]      Busse,  L.  J.,  Miller,  J.  G.,  Yuhas,  D.  E.  , 
Mimbs,  J.  W.,  Weiss,  A.  N.,  and  Sobel  , 
B.  E.,  Phase  Cancellation  Effects:  A 
Source  of  Attenuation  Artifact  Eliminated 
by  a  CdS  Acoustoelectric  Receiver  in  Ultra- 
sound  in  Medicine,  0.  White,  ed.  Vol.  3, 
pp.  1519-1535  (Plenum  Press,  New  York, 
1977). 

[7]      Hueter,  T.  F.,  Visco-Elastic  Losses  in  Tis- 
sues in  the  Ultrasonic  Range,  Wright  Air 
Development  Center.    Wright-Patterson  AFB, 
Ohio,  Tech.  Rept.  No.  57-706,  ASTIA  Doc.  No. 
AD142171  (August,  1968). 

[8]      Carstensen,  E.  L.,  The  Mechanism  of  the 

Absorption  of  Ultrasound  in  Biological  Mate- 
rials, IRE  Trans,  on  Med.  Electronics. 
MET.,  158-162  (1960). 

[9]      Frizzell,  L.  A.  and  Carstensen,  E.  L., 

Ultrasonic  Heating  of  Tissues,  University 
of  Rochester,  Department  of  Electrical 
Engineering  Technical  Report  GM09933-20 
(November,  1975). 

[1^0]    O'Donnell,  M.,  Mimbs,  J.  W.  ,  Sobel,  B.  E.  , 
\    and  Miller,  J.  G.,  Ultrasonic  attenuation 
of  myocardial  tissue:    dependence  on  time 
after  excision  and  on  temperature,  J^. 
Acoust.  Soc.  Am.  (in  press). 

[11]    Carstensen,  E.  L.  and  Schwan,  H.  P., 

Acoustic  properties  of  hemoglobin  solutions, 
J.  Acoust.  Soc.  Am.  31,  205-311  (1959). 

[12]    Kremkau,  F.  W.,  McGraw,  C.  P.,  and  Barnes, 

R.  W. ,  Acoustic  Properties  of  Normal  and  Ab- 
normal Human  Brain  (this  publication,  p.  81) 

[13]  Kishimoto,  T.,  Ultrasonic  absorption  in  , 
bones,  Acustica  8,  179-180  (1959). 

[14]    Dunn,  F.  and  Brady,  J.  K.,  Ultrasonic  ab- 
sorption in  biological  media,  Biofizika 
18,  1063-1066  (1973);  Biophysics  18,  1128- 
1132  (1974). 

[15]    Frizzell,  L.  A.  and  Carstensen,  E.  L., 

Shear  properties  of  mammalian  tissues  at 
low  megahertz  frequencies,  J.  Acoust.  Soc. 
Am.  60_,  1409-1411  (1976). 

[16]  Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of 
absorption  of  ultrasound  in  liver  tissue, 
J.  Acoust.  Soc.  Am.  50,  692-699  (1971). 


35 


[17]    Carstensen,  E.  L.  and  Schwan,  H.  P., 

Absorption  of  sound  arising  from  the  pres- 
ence of  intact  cells  in  blood,  J.  Acoust. 
Soc.  Am.  31,  185-189  (1959). 

[18]    Kremkau,  F.  W.  and  Carstensen,  E.  L., 

Macromolecular  Interaction  in  Sound  Absorp- 
tion, Proc.  of  Workshop  on  Interaction  of 
Ultrasound  and  Biological  Tissues,  Battelle 
Seattle  Research  Center,  Food  and  Drug 
Administration,  U.S.  Dept.  of  Health  Educa- 
tion and  Welfare,  Rockville,  Maryland 
(1972),  pp.  37-42. 

[19]    O'Brien,  W.  D.  and  Dunn,  F.,  Ultrasonic 

examination  of  the  hemoglobin  dissociation 
process  in  aqueous  solutions  of  guanidine 
hydrochloride,  J.  Acoust.  Soc.  Am.  50, 
1213-1215  (1971'r 

[20]    Ponder,  E. ,  The  paracrystall i ne  state  of 
the  rat  red  cell,  J.  Gen.  Physiol.  29, 
89-102  (1945). 

[21]    Kremkau,  F.  W.  and  Carstensen,  E.  L., 

Macromolecular  interaction  in  the  absorp- 
tion of  ultrasound  in  fixed  erythrocytes, 
J.  Acoust.  Soc.  Am.  53,  1448-1451  (1973). 

[22]    Kremkau,  F.  W.  and  Carstensen,  E.  L., 

Macromolecular  Interaction  in  the  Absorp- 
tion of  Ultrasound  in  Biological  Material, 
University  of  Rochester,  Department  of 
Electrical  Engineering  Tech.  Report,  No. 
GM09933-18  (April,  1972). 

[23]    Markham,  J.  J.,  Beyer,  R.  T.,  and  Lindsay, 
R.  B.,  Absorption  of  sound  in  fluids. 
Rev,  of  Mod.  Phys.  22,  353-411  (1951). 

[24]    Eigen,  M.  and  deMaeyer,  L.,  Relaxation 

Methods,  in  Technique  of  Organic  Chemistry, 
A.  Weissberger,  ed. ,  Vol .  8,  Part  II,  pp. 
895-1054  (Wiley,  New  York,  1963). 

[25]    Stuehr,  J.  and  Yeager,  E.,  The  Propagation 
of  Ultrasonic  Waves  in  Electrolytic  Solu- 
tions, in  Physical  Acoustics,  W.  P.  Mason, 
ed..  Vol.  2,  Part  A,  pp.  351-462  (Academic 
Press,  New  York,  1965). 


[26]  Carstensen,  E.  L.,  Relaxation  processes 
in  aqueous  solution  of  MnSO^  and  CoSO^, 
J.  Acoust.  Soc.  Am.  26,  862-864  (1954). 

[27]  Eggers,  F.  and  Funck,  T.,  Ultrasonic  re- 
laxation spectroscopy  in  liquids,  Natur- 
wissenschaften  63,  280-285  (1976). 

[28]    Epstein,  P.  S. ,  On  the  Absorption  of  Sound 
Waves  in  Suspensions  and  Emulsions,  Theo- 
dore von  Karman  Anniversary  Volume,  Cali- 
fornia Institute  of  Technology,  Pasadena, 
California,  1941. 

[29]    Epstein,  P.  S.  and  Carhart,  R.  R. ,  The 
absorption  of  sound  in  suspensions  and 
emulsions.    I.    Water  fog  in  air,  J.  Acoust. 
Soc.  Am.  25,  553-565  (1953) 

[30]    Urick,  R.  J.,  The  absorption  of  sound  in 
suspensions  of  irregular  particles,  J_. 
Acoust.  Soc.  Am.  20,  283-289  (1948). 

[31]    Fry,  W.  J.,  Mechanism  of  acoustic  absorp- 
tion in  tissue,  J.  Acoust.  Soc.  Am.  24, 
412-415  (1952). 

[32]    Allegra,  J.  B.  and  Hawley,  S.  A.,  Attenua- 
tion of  sound  in  suspensions  and  emulsions: 
theory  and  experiments,  J.  Acoust.  Soc. 
Am.  51,  1545-1564  (1972). 

[33]    Ahuja,  A.  J.,  Acoustical  properties  of 
blood:    a  look  at  the  basic  assumptions, 
Med.  Phys.  1,  311  (1974). 

[34]  O'Donnell,  M.  and  Miller,  J.  G.,  Mecha- 
nisms of  Ultrasonic  Attenuation  in  Soft 
Tissue  (this  publication,  p.  37). 

[35]    Kessler,  L.  W. ,  VHF  ultrasonic  attenuation 
in  mammalian  tissue,  J.    Acoust.  Soc.  Am. 
53,  1759-1760  (1973). 

[36]    Hawley,  S.  A.  and  Dunn,  F.,  Ultrasonic 

Absorption  in  Aqueous  Solutions  of  Dextran, 
J.  Chem.  Phys.  50,  3523-3526  (1969). 


36 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ .   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


MECHANISMS  OF  ULTRASONIC  ATTENUATION  IN  SOFT  TISSUE 


M.  O'Donnell  and  J.  G.  Miller 

Laboratory  for  Ultrasonics 
Department  of  Physics 
Washington  University 
St.  Louis,  Missouri    63130,  U.S.A. 


Ultrasonic  loss  mechanisms  which  arise  from  the  microscopically  i nhomogeneous  nature 
of  soft  tissue  are  investigated  over  the  frequency  range     1  to     10  MHz.  Contribu- 
tions to  the  ultrasonic  attenuation  due  to  viscous  relative  motion  losses  and  thermal 
losses  are  shown  to  exhibit  an  approximately  linear  dependence  on  frequency.  Numerical 
estimates  of  the  attenuation  arising  from  these  mechanisms  are  compared  with  the  re- 
sults of  experiments  for  a  representative  substance  in  each  of  three  attenuation  cate- 
gories:   low  (blood),  medium  (heart),  and  high  (skin).    Inhomogeneity  losses  can  account 
for     60  percent  of  the  attenuation  observed  in  heart  and  skin,  and  thus  may  contribute 
a  non- negl igi bl e  fraction  to  the  attenuation  observed  in  most  soft  tissue. 

Key  words:    Inhomogenei ties ;  inhomogenity  thermal  losses;  mechanisms;  viscous  relative 
motion . 


In  efforts  to  account  for  the  ultrasonic  at- 
tenuation exhibited  by  soft  tissue,  two  types 
of  mechanisms  have  been  considered:    (1)  struc- 
tural or  chemical  relaxation  of  the  macro- 
molecular  constituents  [1-3]^,  and  (2)  proc- 
esses arising  from  the  i nhomogeneous  nature  of 
tissue,  including  scattering,  viscous  relative 
motion,  and  thermal  losses  [1,3-8].    In  this 
paper  we  estimate  the  fraction  of  the  total  at- 
tenuation observed  in  the  1  to  10  MHz  frequency 
range  which  is  attributable  to  processes  arising 
solely  from  the  microscopically  i nhomogeneous 
nature  of  soft  tissue,  as  opposed  to  that  frac- 
tion presumed  to  arise  from  macromol ecu! ar  re- 
laxation processes. 

We  consider  a  model  in  which  a  longitudinal 
ultrasonic  wave  propagates  through  tissue  which 
is  viewed  as  a  suspension  of  scatterers  in  a 
liquid  medium  (cytoplasm).    For  the  sake  of 
simplicity  all  scatterers  are  considered  to  be 
spherical  in  shape.    At  an  interface  between 
the  medium  and  a  scatterer,  longitudinal,  vis- 
cous, and  thermal  waves  may  be  excited  in  the 
suspending  medium  due  to  the  discontinuity  of 
acoustic  and  thermal  properties  at  the  surface 
1    of  the  scatterer.    Energy  that  is  coupled  into 
!]    these  waves  is  lost  from  the  incident  ultra- 
i    sonic  beam  and  thus  contributes  to  the  total 
1    attenuation  that  is  observed. 

For  scatterers  small  compared  to  the  wave- 
[j    length  of  the  incident  wave  the  reradiated 
1,  longitudinal  wave  corresponds  to  Rayleigh 
j   scattering.    The  effect  contributes  a  term  to 
■   the  attenuation  coefficient  that  varies  as 
the  fourth  power  of  frequency  and  the  third 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


power  of  the  scatterer  radius  for  a  fixed 
volume  concentration  of  scatterers.  The 
magnitude  of  Rayleigh  scattering  depends  upon 
the  square  of  the  difference  in  adiabatic 
compressibilities  of  the  scatterer  and  the 
suspending  medium  and  upon  the  square  of  the 
difference  in  densities.    The  numerical  value 
of  the  contribution  due  to  Rayleigh  scatter- 
ing is  several  orders  of  magnitude  lower  than 
the  observed  attenuation  coefficient  of  soft 
tissue  in  the  frequency  range  1  to  10  MHz  [1,2]. 

Details  of  the  scattering  process  for  scat- 
terers of  dimensions  which  are  not  small  com- 
pared to  the  ultrasonic  wavelength  are  less  well 
understood.    (Specular  reflections  from  objects 
much  larger  in  size  than  the  ultrasonic  wave- 
length are  explicitly  excluded  here.)  The 
general  features  of  the  scattering  problem  in- 
dicate that  the  magnitude  of  scattering  events 
depends  primarily  on  the  square  of  the  differ- 
ences in  densities  and  compressibilities  of  the 
scatterer  and  the  suspending  medium.    In  soft 
tissue,  scatterers  of  dimensions  comparable  to 
ultrasonic  wavelengths  in  the  1  to  10  MHz  range 
(i.e.,  scatterers  of  sizes  ranging  from  0.05 
mm  to  1  mm)  exhibit  densities  and  compressi- 
bilities yery  close  to  those  of  the  suspending 
medium.    Thus  losses  arising  from  these  scat- 
tering events  are  expected  to  represent  small 
contributions  to  the  attenuation  coefficient. 
Such  scattering  events,  however,  may  be  of 
central  importance  in  scattering  and  reflec- 
tion (i.e.,  180°  backscattering)  experiments. 

Viscous  drag  losses  arise  from  the  genera- 
tion of  a  highly  damped  viscous  wave  in  the 
suspending  medium.    The  scatterers  attempt  to 
mirror  the  motion  of  the  suspending  medium  to 
all  extent  determined  by  a  function  which  de- 
pends upon  the  relative  densities  of  the  scat- 


37 


terers  and  the  medium,  the  ultrasonic  fre- 
quency, and  the  viscosity  of  the  medium  [4,7]. 
For  fixed  frequency,  viscosity,  and  volume 
concentration  of  scatterers,  the  contribution 
of  viscous  drag  losses  to  the  attenuation  co- 
efficient is  predicted  to  rise,  reach  a  maxi- 
mum, and  subsequently  fall  as  a  function  of 
scatterer  size.    A  similar  rise,  plateau,  and 
fall  is  predicted  if  viscous  relative  motion 
losses  are  plotted  as  a  function  of  frequency 
for  fixed  scatterer  size.    Contributions  to  the 
attenuation  of  ultrasound  in  tissue  by  viscous 
relative  motion  losses  were  considered  by  Fry 
[6],  Carstensen  and  Schwan  [1],  and  Kremkau, 
Carstensen,  and  Aldridge  [3]. 

An  additional  contribution  to  the  ultrasonic 
attenuation  arises  from  the  generation  of  a 
highly  damped  thermal  wave  at  the  surface  of  a 
scatterer.    (Strictly  speaking,  the  viscous  drag 
and  thermal  effects  are  coupled,  but  the  strength 
of  this  coupling  is  sufficiently  small  that  it 
can  be  ignored  [4].)    The  magnitude  of  the  losses 
associated  with  the  generation  of  the  thermal 
wave  depends  upon  frequency  and  upon  the  relative 
heat  capacities,  coefficients  of  thermal  expan- 
sion, and  thermal  conductivities  of  the  scat- 
terers and  suspending  medium.    At  low  frequencies 
the  system  is  in  an  isothermal  limit  in  which 
energy  is  reversibly  exchanged  between  the  car- 
rier medium  and  a  scatterer,  which  closely  follow 
each  other  in  temperature  ("isothermal"),  so  that 
relatively  little  loss  of  energy  occurs.    At  high 
frequencies  the  system  is  in  an  adiabatic  limit 
in  which  temperature  changes  in  the  suspending 
medium  occur  so  rapidly  that  they  cannot  be  com- 
municated to  a  scatterer.    Thus  little  heat 
transfer  occurs  ("adiabatic")  and  relatively 
little  energy  is  dissipated.    Maximum  energy  is 
dissipated  in  the  intermediate  range  of  fre- 
quencies where  the  thermal  response  of  the  scat- 
terers to  the  incident  compressi onal  ultrasonic 
wave  lags  sufficiently  behind  that  of  the  sus- 
pending medium  that  substantial  energy  is  dis- 
sipated.   Thermal  losses  in  i nhomogeneous  media 
have  been  considered  by  Epstein  and  Carhart  [4], 
Allegra  and  Hawley  [7],  Kremkau,  Carstensen,  and 
Aldridge  [3],  and  Ahuja  [8]. 

In  order  to  estimate  the  contribution  to  the 
observed  attenuation  arising  from  viscous  and 
thermal  effects,  it  is  necessary  to  identify  the 
relevant  scatterers  in  biological  specimens.  In 
principle,  all  inhomogeneities  contribute  to  the 
observed  attenuation  at  any  frequency.  The 
physical  arguments  presented  above,  however, 
suggest  that  for  a  fixed  range  of  frequencies 
only  scatterers  exhibiting  specific  physical 
properties  that  define  the  relatively  narrow 
ranges  over  which  either  viscous  or  thermal 
losses  are  near  maximum  need  be  considered. 
Numerical  computations  discussed  below  indicate 
that  for  the  range  of  scatterer  dimensions 
which  results  in  significant  viscous  relative 
motion  losses  or  thermal  losses,  contributions 
to  the  attenuation  coefficient  exhibit  an  ap- 
proximately linear  dependence  on  frequency, 
i.e.,  log  (attenuation)  versus  log  (frequency) 
exhibits  a  slope  of  0.8  to  1.2  over  the  fre- 
quency range  1  to  10  MHz.    Therefore,  the  slope 
of  the  attenuation  versus  frequency  curve  over 
the  range  of  1  to  10  MHz  (hereafter  referred  to 
as  the  slope  of  the  attenuation)  can  serve  as  a 
useful  index  of  ultrasonic  properties.  This 


conclusion,  arrived  at  from  theoretical  con- 
siderations, is  consistent  with  the  results  of 
experiments,  which  indicate  an  approximately 
linear  dependence  on  frequency  for  the  ultra- 
sonic attenuation  coefficient  of  soft  tissue. 
The  slope  of  the  attenuation  represents  a  de- 
sirable index  since  in  transmission  experiments 
it  is  less  susceptible  to  errors  associated 
with  impedance  discontinuities  and  may  be  less 
susceptible  to  errors  resulting  from  phase 
cancellation  effects  than  is  the  apparent  at- 
tenuation coefficient. 


LO 

U.8 

/    \              THERMAL  ATTENUATION 

'e 

/       \                   Con^  =  17% 

b 

0.6 

/        \               Pscatt  =  1 4lg/cm3 

Its  of 

0.4 

c 

0.2 

SLOPE 

0.0 

1     1       .  1  

0.01  01  I  10  100 

SCATTERER  DIAMETER  ( /xm  ) 

Fig.  1.    Slope  of  the  attenuation  versus  frequency 
curve  over  the  range  1  to  10  MHz  arising 
from  thermal  losses  plotted  as  a  function 
of  scatterer  diameter. 

In  figure  1  the  contribution  to  the  slope  of 
the  attenuation  arising  from  thermal  losses  is 
plotted  as  a  function  of  scatterer  size.  The 
physical  properties  of  the  scatterers  and  the 
suspending  medium  used  in  calculating  the  re- 
sults shown  in  figure  1  appear  in  table  1  which 
is  discussed  below.    As  illustrated  in  figure  1, 
contributions  from  thermal  losses  are  largest 
when  the  thermal  wavelength  Xi  of  the  suspending 
medium  approximately  equals  2  ir  times  the  radius 
of  the  scatterer.    (A^  is  on  the  order  of  0.4  to 
1.3  m  in  the  frequency  range  1  to  10  MHz.) 

The  slope  of  the  attenuation  which  results 
from  the  coupling  of  energy  into  a  viscous  wave 
is  plotted  as  a  function  of  scatterer  size  in 
figure  2.    As  illustrated,  viscous  relative  mo- 
tion losses  are  largest  when  the  viscous  wave- 
length Ay  of  the  suspending  medium  approximate- 
ly equals  2  ir  times  the  radius  of  the  scatterers. 
(Ay  is  the  order  of  1.8  to  5.5  vim  in  the  1  to 
10  MHz  frequency  range. ) 

Our  calculations  indicate  that  viscous  rela- 
tive motion  losses  dominate  thermal  losses  in 
the  1  to  10  MHz  frequency  range  for  a  wide 
range  of  scatterer  properties  consistent  with 
the  microscopic  anatomy  of  soft  tissue.  We 
thus  focus  on  viscous  relative  motion  losses  in 
an  attempt  to  identify  specific  physical  proper- 
ties that  must  be  exhibited  by  the  relevant 
scatterers.    The  numerical  magnitude  of  the 
viscous  relative  motion  term  is  approximately 
proportional  to  (p'/p  -  1)^,  where  p'  is  the 
density  of  the  scatterer  and  p  is  the  density 
of  the  suspending  medium.    Therefore,  signifi- 
cant contributions  to  the  observed  attenuation 
result  only  from  those  scatterers  which  (1)  ex- 
hibit sizes  that  maximize  viscous  relative  mo- 


38 


75 


X  60 


45 


30 


/~\              VISCOUS  ATTENUATION 

/        \                    Con^,  =  17% 

/           \        '^suspending  =  2.5cp 

/            \  medium 

- 

- 

/               \             ^scQtt  =  I.41g/cm3 
\                    ■       -  ' 

-    ,     1  1 

111       1  1  

0.01  0.1  I  10 

SCATTERER  DIAMETER  (/xm) 


100 


Fig.  2.    Slope  of  the  attenuation  (1  to  10  MHz) 
arising  from  viscous  relative  motion 
losses  plotted  as  a  function  of  scatterer 
diameter. 

tion  losses,  (2)  have  densities  significantly 
different  from  that  of  the  suspending  medium, 
and  (3)  are  present  in  sufficient  concentra- 
tions.   Aggregates  of  structural  protein  such 
as  fibrils  of  collagen  or  muscle  myofibrils  ap- 
pear to  be  the  principal  constituents  of  soft 
tissue  which  possess  these  properties.  (Dunn 
and  his  colleagues  have  identified  collagen  as 
a  significant  determinant  of  the  ultrasonic 
properties  of  soft  tissue,  in  part  because  of 
its  large  compressional  modulus  and  ubiquity 
[9,10].) 

Limiting  our  attenuation  to  scatterers  ex- 
hibiting the  requisite  physical  properties,  we 
estimate  the  contribution  to  the  observed  at- 
tenuation arising  from  microscopic  inhomo- 
geneities.    Specific  numerical  values  for  the 
properties  of  the  scatterers  and  the  suspend- 
ing medium  which  were  used  in  our  calculations 
are  presented  in  table  1.    An  attempt  has  been 
made  to  estimate  the  contribution  to  the  ex- 
perimentally observed  attenuation  arising  only 
from  viscous  and  thermal  effects  for  a  repre- 
sentative substance  in  each  of  three  attenua- 
tion categories:    low  (blood),  medium  (heart). 


and  high  (skin).    Results  of  these  calculations 
are  summarized  in  table  2.    Losses  associated 
with  viscous  relative  motion  are  seen  to  domi- 
nate thermal  losses  in  the  1  to  10  MHz  range, 
a  feature  which  appears  to  be  preserved  over  a 
wide  range  of  scatterer  properties. 

The  quantitative  results  presented  in  table 
2  indicate  that  losses  arising  solely  from 
microscopic  inhomogeneities  appear  to  account 
for  a  significant  fraction  ('v  60  percent)  of 
the  total  observed  attenuation  of  heart  and 
skin.    These  results  suggest  that  i nhomogeneity 
losses  may  contribute  a  non-negligible  fraction 
to  the  attenuation  of  most  soft  tissue.  Al- 
though the  exact  magnitude  of  these  losses  is 
certain  to  be  altered  by  the  use  of  more  real- 
istic assumptions  (e^. ,  non-spherical  scat- 
terers and  improved  estimates  of  the  values  of 
physical  properties  specified  in  table  1),  the 
general  conclusions  of  this  study  would  not  be 
altered  if  the  illustrative  numerical  results 
presented  in  table  2  were  in  error  by  30  percent. 


Table  2.    Slope  of  attentuation  coeff icient-vs-frequency 
(cm-iMHz-i 

Tissue       Viscous  Thermal  Theoretical  Experimental  Percent 

value^  value       accounted  forC 


Blood 
(40%  Hct) 
Heart  0.042 
Skin  0.101 


0.003  0.000 


0.001 
0.001 


0.003 


0.043 
0.102 


0.03ld 


10 


0.0726  60 
0.17of  60 


Numerical  results  presented  to  3  significant  digits  for 
j^i  1 1  ustrati ve  purposes  only. 
^Sum  of  contributions  from  columns  2  and  3. 

Percent  of  experimentally  observed  values  (column  5)  accounted 

for  by  sum  of  contributions  (column  4)  arising  from  losses  due 

.to  microscopic  inhomogeneities. 

Carstensen,  E.  L.,  Li,  K. ,  and  Schwan,  H.  P.,  J.  Acoust.  Soc. 
Am.  25,  286-289  (1953). 

^O'Donnel,  H.,  Mimbs,  J.  W. ,  Sobel ,  B.  E.,  and  Miller  J.  G., 
Ultrasonic  Attenuation  in  Normal  and  Ischemic  Myocardium 

^(this  conference). 
Dussik,  K.  T.,  Kyriazidov,  M.,  Fritch,  0.  J.,  and  Sear,  R.  S., 
Am.  J.  Rhys.  Med.  37,  160  (1958). 


Table  1.    Acoustic  and  thermal  properties  of  tissue  scatterers  (T  =  20  'C). 


Tissue 

Scatterers 

Size  range^ 

Density 

Vol ume 

Thermal 

Coefficient 

Speci  f ic 

concentration  conductivity 

of  thermal 

heat 

expansion 

%  of  total 

10-5 

(pm) 

(g/cm3) 

vol ume 

cal/s-cm-^C 

10-"  deg-i 

cal /g-°C 

Blood 

Red  cells 

4.6  -  5. 6b 

1 .09b 

40b 

1.1b 

1.2c 

0.8b 

(40%  Hct) 

1  .32^ 

Heart 

Muscl e 

1,0  -  2.0^ 

18d 

0.4c 

1 .2C 

0.4c 

myof i  bri 1 s 

1  .4lf 

Skin 

Ccl lagen 

0.7  -  1.56 

259 

0.4c 

1 .2= 

0.4c 

f ibri 1 s 

1  .03b 

Suspendi  ng^ 

1 .5b 

1.41 

1  .Ob 

— medium 


.Uniform  distribution  of  scatterers  with  diameters  expressed  in  micrometers. 
°Ahuja,  A.  S.,  Med.  Phj^.  1,  311-316  (1974). 
^Estimated . 

Bailey,  Kenneth,  Proteins ,  H.  Neurath  and  Kenneth  Bailey,  eds.,  Chapt.  4  (Academic  Press, 

New  York,  1954). 
.pGross,  Jerome,  Scientific  American,  120-130  (May  1961). 

Pomeroy,  C.  D.  and  Milton,  R.  J.,  J.  Soc.  Leather  Trades  Chem.  35,  360  (1951). 
^Chupvil ,  .M. ,  Physiology  of  Connective  Tissue  (Butterworth,  London,  1967). 
.Viscosity  (1  to  10  MHz)  estimated  to  be  2.5  centipoise. 

Epstein,  P.  S.  and  Carhart,  R.  C,  J.  Acoust.  Soc.  Am.  25,  553  (  1953). 


39 


Acknowl edgment 


This  work  was  supported  in  part  by  grants 
HL19537,  HL17646,  and  HL07081  from  the  National 
Institutes  of  Health. 

Pranoat  Suntharothok-Priesmeyer  was  respon- 
sible for  production  of  the  text  and  illustra- 
tions. 


References 

[1]     Carstensen,  E.  L.  and  Schwan,  H.  P., 
Absorption  of  sound  arising  from  the 
presence  of  intact  cells  in  blood,  J_. 
Acoust.  Soc.  Am.  31_,  185  (1959). 

[2]  Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of 
absorption  of  ultrasound  in  liver  tissue, 
J.  Acoust.  Soc.  Am.  50,  692  (1971). 

[3]      Kremkau,  F.  W. ,  Carstensen,  E.  L.,  and 
Aldridge,  W.  G.,  Macromolecular  inter- 
actions in  the  absorption  of  ultrasound 
in  fixed  erythrocytes,  J.  Acoust.  Soc. 
Am.  53,  1448  (1973). 

[4]     Epstein,  P.  S.  and  Carhart,  R.  C,  The 
absorption  of  sound  in  suspensions  and 
emulsions.    I.    Water  fog  in  air,  J_. 
Acoust.  Soc.  Am.  25.,  553  (1953). 

[5]     Urick,  R.  J.,  The  absorption  of  sound  in 
suspensions  of  irregular  particles,  J_. 
Acoust.  Soc.  Am.  20,  283  (1948). 

[6]     Fry,  W.  J.,  Mechanism  of  acoustic  absorp- 
tion in  tissue,  J.  Acoust.  Soc.  Am.  24 
412  (1952). 

[7]  Allegra,  J.  R.  and  Hawley,  S.  A.,  Attenua- 
tion of  sound  in  suspensions  and  emulsions 
theory  and  experiments,  J.  Acoust.  Soc.  Am 
51,  1545  (1972). 

[8]     Ahuja,  A.  J.,  Acoustical  properties  of 
blood:    a  look  at  the  basic  assumptions, 
Med.  Phys.  U  311  (1974). 

[9]     Fields,  S.  and  Dunn,  F.,  Correlation  of 
echographic  visualization  of  tissue  with 
biological  composition  and  physiological 
state,  J.  Acoust.  Soc.  Am.  54,  809  (1973). 

[10]    O'Brien,  W.  D.  Jr.,  The  Role  of  Collagen 
in  Determining  Ultrasonic  Propagation 
Properties  in  Tissue,  in  Acoustical  Holo- 
graphy, L.  W.  Kessler,  ed. ,  Vol .  7  (Plenum 
Press,  New  York,  1976). 


CHAPTER  3 
ATTENUATION  AND  VELOCITY  II: 
METHODOLOGY  AND  MEASUREMENTS 


41 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ELEMENTS  OF  TISSUE  CHARACTERIZATION 
Part  II.    Ultrasonic  Propagation  Parameter  Measurements 


S.  A.  Goss,  R.  L.  Johnston,  V.  Maynard,  L.  Nider, 
L.  A.  Frizzell,  W.  D.  O'Brien,  Jr.,  and  F.  Dunn 

Bioacoustics  Research  Laboratory 
University  of  Illinois 
Urbana,  Illinois    61801,  U.S.A. 


Methods  employed  at  the  Bioacoustics  Research  Laboratory  of  the  University  of 
Illinois  for  the  determination  of  ultrasonic  propagation  properties  of  biological 
media  are  described,  with  attention  devoted  to  attenuation,  absorption  and  veloci- 
ty measurements  of  both  longitudinal  and  shear  ultrasonic  waves.    These  include 
systems  specifically  for  the  ultrasonic  characterization  of  soft  tissues  and  for 
solutions  of  biologically  significant  macromolecules.    Each  method  is  presented 
in  terms  of  theory,  limitations,  applicability,  and  possible  sources  of  error. 
Important  new  techniques  from  other  laboratories  are  also  discussed. 

Key  words:    Ultrasonic  absorption;  ultrasonic  attenuation;  ultrasonic  instrumenta- 
tion; ultrasonic  measurements;  ultrasonic  spectroscopy;  ultrasonic 
tissue  characterization;  ultrasonic  tissue  parameters;  ultrasonic 
tissue  signature;  ultrasonic  velocity. 


1.  Introduction 

A  number  of  measuring  techniques  have  been 
developed,  or  are  otherwise  employed,  at  the 
Bioacoustics  Research  Laboratory  of  the  Uni- 
versity of  Illinois  in  the  research  efforts  as- 
sociated with  the  propagation  characteristics  of 
ultrasound  in  biological  tissue,  and  toward  a 
basic  understanding  of  the  mechanism(s)  re- 
sponsible for  the  acoustic  properties  exhibited 
by  those  tissues.    These  include  specialized 
systems  for  biological  liquids  and  for  soft 
tissues  which  are  capable  of  measuring  attenua- 
tion, absorption,  and  velocity  of  longitudinal 
ultrasonic  waves  as  functions  of  frequency,  tem- 
perature and,  where  appropriate,  pH  and  ambient 
pressure.    In  addition,  a  system  is  presently 
being  developed  to  measure  the  shear  acoustical 
properties  of  biological  materials  as  a  function 
of  frequency  and  temperature.    These  techniques, 
thirteen  in  all,  are  described  in  this  paper  in 
terms  of  parameters  measured,  theory,  applica- 
tion, precision/accuracy,  frequency,  and  tempera- 
ture range,  so  that  each  technique  may  be  as- 
sessed separately. 

2.    Attenuation  and  Velocity 
Measurements  in  Soft  Tissue 

Attenuation  of  longitudinal  waves  in  soft  tis- 
sue specimens  are  determined  using  the  radiation 
force,  pulse  transmission,  and  standing  wave 
techniques.    Velocity  measurements  in  soft  tis- 
sues can  be  made  by  observing  the  time  of  flight 
of  an  acoustic  pulse  through  a  known  path  length 
of  a  tissue  sample,  and  by  acoustic  interferometry. 


A.    Attenuation:    Radiation  Force  Method 

The  phenomenon  of  radiation  force  provides  a 
primary  method  for  the  measurement  of  the  second 
order  quantities  of  intensity  and  power  [l-4]i. 
Radiation  force  is  a  direct  result  of  energy 
transport  by  the  sound  wave,  and  is  equal  to  the 
time  rate  of  change  of  momentum  of  the  wave. 
Thus  a  continuous  wave  incident  on  a  reflecting 
or  absorbing  object  will  produce  a  time  inde- 
pendent force  on  that  object  equal  to,  and  in 
the  direction  of,  the  time  rate  of  change  of 
momentum.    Consequently,  a  sensitive  balance  can 
be  employed  to  measure  the  radiation  force  exert- 
ed by  an  ultrasonic  beam  incident  on  a  suspended 
target.    It  then  follows  that  force,  measured  as 
a  change  in  tension  of  the  suspending  wire,  on  a 
perfectly  absorbing  target  intercepting  a  verti- 
cally directed  sound  beam  is 


where  F  is  the  measured  change  in  tension  of  the 
suspending  wire,  P  is  the  time  average  power  in- 
tercepted by  the  target,  and  c  is  the  velocity 
of  propagation.    Thus,  in  water,  1  milliwatt  of 
incident  power  will  exert  a  force  equivalent  to  a 
weight  of  67  pg,  which  is  measured  as  an  apparent 
target  weight  change.    Similarly,  a  perfectly 
reflecting  target  intercepting  a  vertically 
directed  sound  beam  will  produce  a  measurable 
change  in  suspension  tension,  given  as 


Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


43 


F=2Pcos29 
c 

where  e  is  the  angle  between  the  beam  axis  and  the 
normal  to  the  target  surface.    Consequently  in 
water,  when  the  target  is  inclined  45°  to  the  ver- 
tical, the  sensitivity  of  the  balance  is  again  67 
pg/mW  of  intercepted  power.    Thus,  a  phase  insensi- 
tive, frequency  independent  technique  for  power  and 
intensity  attenuation  measurements  becomes  available 
by  interposing  a  tissue  sample  between  the  sound 
source  and  the  sound  intercepting  target,  by  deter- 
mining that  power  lost  or  redirected  with  the  tissue 
sample  in  place  compared  with  measured  power  with 
the  tissue  absent  [2,5,6].    Care  must  be  taken  to 
insure  that  the  system's  sensitivity  is  actually  that 
that  predicted  for  such  idealized  cases.    The  atten- 
uation present  in  the  liquid  media  (water  or  physio- 
logical saline)  will  inevitably  cause  acoustic 
streaming  which  will  lead  to  systematic  errors  in 
the  power  estimation.    The  streaming  effect  can  be 
minimized  by  positioning  an  acoustically  transparent 
barrier  of,  for  example,  stretched  polyethylene, 
directly  in  front  of  the  target. 

A  second  possible  source  of  error  is  due  to  the 
inherent  insensitivity  of  the  measurement  system 
to  horizontal  forces.    For  maximum  accuracy,  the 
main  lobe  of  the  sound  beam  should  be  directed 
precisely  vertically.    Because  the  beam  may  con- 
tain components  of  momentum  in  the  plane  perpen- 
dicular to  the  main  lobe,  the  radiation  force 
balance  method  does  not  respond  to  all  the  energy 
emitted  by  the  source.    As  the  temperature  of  the 
target  increases  due  to  the  fact  that  energy  is 
being  deposited  in  it,  thermal  expansion  of  the 
target  can  introduce  an  error  by  virtue  of  in- 
creasing its  buoyancy  and  resulting  in  an  apparent 
target  weight  change.    The  ultimate  accuracy  of 
the  radiation  force  balance  method  is  limited  by 
noise  associated  with  mechanical  vibrations  and 
atmospheric  perturbations,  which  can  be  reduced  by 
proper  design  considerations.    However,  in  the 
presence  of  pressure  gradients,  the  target  is 
acted  on  by  forces  proportional  to  the  target 
volume,  making  a  small  target  advantageous. 

B.    Attenuation:    Pulse  Transmission  Method 

The  pulse  transmission  technique  is  applicable 
to  measurements  of  sound  pressure  attenuation  in 
various  types  of  biological  specimens  [7,8].  The 
instrumentation  consists  of  a  transmitting  ultra- 
sonic transducer  immersed  in  a  bath  (usually 
physiological  saline)  which  serves  as  the  acoustic 
coupling  medium  and  aligned  axially  with  either  a 
reflecting  surface  or  a  receiving  transducer.  The 
associated  electronic  instrumentation  provides  an 
RF  pulse  to  the  transmitting  crystal  and  also  trig- 
gers the  sweep  of  an  oscilloscope  a  variable,  but 
selected,  length  of  time  after  the  initiation  of 
the  transmitting  pulse.    An  oscilloscope  serves 
to  display  the  amplified  received  signal.  Measure- 
ment of  sound  attenuation  can  be  performed  as  fol- 
lows.   First,  with  no  sample  in  the  ultrasonic 
path,  the  gain  of  the  receiving  amplifier  is  ad- 
justed to  give  a  signal  display  of  a  predetermined 
amplitude.    Then,  with  the  specimen  in  place,  the 
gain  of  the  system  is  adjusted,  either  by  increas- 
ing the  receiving  amplifier  gain  or  by  removing 
electrical  attenuation  in  the  signal  path,  to  com- 
pensate for  the  loss  of  acoustic  energy  in  the 
sample  to  bring  the  display  trace  to  the  same 


preselected  amplitude.    The  attenuation  measure- 
ment is  repeated  for  other  samples  of  the  same 
material,  but  of  varying  thicknesses.    The  mea- 
sured attenuation  values  are  then  plotted  versus 
sample  thickness,  and  the  slope  of  the  resulting 
straight  line  yields  the  attenuation  per  unit 
length  of  the  specimen. 

C.  Velocity:    Pulse  Transit  Time  Method 

Velocity  measurements  in  tissue  can  be  made 
simply  by  measuring  the  time  of  flight  of  an 
acoustic  pulse  over  a  known  path  length  of  the 
specimen  [9,10].    Differential  techniques  should 
be  used  to  minimize  uncertainties  in  path  length 
and  t^'me  delay  measurements,  such  as  the  uncertain- 
ty of  the  actual  location  of  the  active  face  of 
the  transducer.    If  commercial  transducers  are 
employed,  additional  allowances  must  be  made  for 
the  quarter-wave  impedance  matching  layer  on  their 
surface.    Errors  may  be  encountered  in  time  of 
flight  measurements  of  the  acoustic  signal  in  in- 
homogeneous  media,  such  as  tissues.    Since  a  short 
duration  acoustic  pulse  contains  a  broad  spectrum 
of  components,  the  frequency  dependent  effects  of 
velocity  dispersion,  attenuation,  and  multiple 
phase  shifts  at  tissue  interfaces,  can  distort 
and  delay  the  signal  [11].    Techniques  which 
rely  on  the  detection  of  the  leading  edge  of  the 
received  acoustic  pulse  for  timing  are  prone  to 
the  greatest  inaccuracies  under  such  distortions. 

D.  Velocity:    Acoustic  Interferometric 

Method 

Nearly  an  order  of  magnitude  improvement  in  ac- 
curacy can  be  realized  by  the  use  of  acoustic  in- 
terferometric techniques  employing  CW  excitation 
[12,13].    By  sandwiching  the  tissue  sample  between 
transmitting  and  receiving  transducers,  or  between 
a  transducer  and  a  reflecting  target,  a  standing 
wave  is  created  in  the  tissue  sample.    It  has  been 
shown  that  tissue  may  be  distorted  in  this  manner 
up  to  25  percent  without  affecting  the  measured 
velocity  [12].    By  monitoring  the  maxima  and 
minima  of  the  received  acoustic  signal  as  the  path 
length  is  changed,  the  standing  wave  pattern  and 
the  wavelength  of  sound  in  the  tissue  can  be  mea- 
sured.   This  information  combined  with  the  fre- 
quency employed  for  the  measurement  yields  the 
velocity  of  sound  in  the  sample.  Alternatively, 
the  path  length  may  be  held  fixed  and  the  frequen- 
cy of  excitation  changed.    The  change  in  frequency 
necessary  to  move  from  one  mode  in  the  standing- 
wave  pattern  to  the  next  is  directly  proportional 
to  velocity.    The  accuracy  of  such  interferometric 
techniques,  under  tight  temperature  control,  can 
reach  about  0.1  percent. 

E.  Attenuation  and  Velocity:  Standing 

Wave  Method 

While  the  measuring  techniques  described  above 
may  be  applied  to  the  measurement  of  the  ultra- 
sonic absorption  and  velocity  of  most  biological 
tissues,  some  specimens  require  specialized 
techniques . 

Lung,  by  virtue  of  its  extraordinarily  high 
attenuation,  presents  some  problems  in  determina- 
tion of  its  ultrasonic  propagation  properties. 
The  following  method  has  been  developed  and  yields 
results  in  agreement  with  other  similar  ones  [8]. 


44 


Briefly,  the  lung  is  suspended  in  the  sound  field 
between  the  sound  source  and  an  absorption  chamber, 
to  eliminate  standing  waves  beyond  the  lung.  The 
transient  thermoelectric  probe  [14]  provides  a 
convenient  acoustic  detector  to  investigate  (A) 
the  acoustic  field  between  the  specimen  and  the 
source,  to  determine  the  axial  standing  wave  pat- 
tern, and  (B)  the  acoustic  field  between  the  speci- 
men and  the  absorption  chamber,  to  determine  the 
wave  amplitude  transmitted  beyond  the  lung  speci- 
men   15    (see  fig.  1).    Probing  the  field  in  A 
yields  the  fraction  of  incident  energy  reflected 
at  the  lung-saline  interface.    Assuming  that  in- 
finitesimal wave  acoustics  prevails  and  that  the 
attenuation  in  the  lung  is  sufficiently  great  that 
multiple  reflections  within  the  specimen  need  not 
be  considered,  the  speed  of  sound  in  the  lung  can 
be  obtained  from 


!pv)i 


ung 


(pv)saline 


SWR 


(3) 


where  the  p's  are  the  known  densities  and  SWR  is  the 
observed  standing  wave  ratio  (between  the  source  and 
specimen)  and  is  greater  than  unity.    The  attenua- 
tion coefficient  per  unit  path  length  is  determined 
from  a  knowledge  of  the  energy  reflected  at  the  two 
lung-saline  interfaces,  the  thickness  of  the  sample, 
and  the  acoustic  intensity  detected  by  the  probe  in 
B  in  accordance  with  the  relation 


Ine 


-2al 


(4) 


where  Iq  and  Ij  are,  respectively,  the  acoustic  in- 
tensities at  the  lung-saline  interface  nearest  to 
and  farthest  from  the  source,  1  is  the  thickness 
of  the  sample,  and  a  is  the  attenuation  coefficient. 


3. 


Absorption  Measurements 
in  Soft  Tissue 


The  transient  thermoelectric  method  is  well 
suited  for  determining  acoustic  absorption  in 


small  volumes  of  highly  absorbing  liquid  or  tis- 
sue ir[  vivo  as  well  as  i_n  vitro,  and  may  be  the 
only  method,  applicable  to  tissues,  that  mea- 
sures directly  absorption  rather  than  attenua- 
tion.   As  such  it  is  invaluable  in  determining 
the  portion  of  attenuation  due  to  absorption  vis- 
a-vis that  portion  due  to  scattering  effects 
L16,17].    The  method  requires  that  a  thermocouple 
junction  of  small  diameter  relative  to  the  wave- 
length be  implanted  in  the  sample.    The  thermo- 
couple and  surrounding  sample  are  then  exposed  to 
a  plane  traveling  wave  ultrasonic  field  having  a 
temporally  rectangular  envelope  of  known  intensity. 
The  typical  thermoelectric  emf  response  to  a  1 
second  ultrasonic  pulse  has  an  initial  fast  rise 
which  results  from  conversion  of  acoustic  energy 
into  heat  by  the  viscous  forces  acting  between 
the  thermocouple  wire  and  the  sample.    This  por- 
tion of  the  response  approaches  equilibrium  very 
quickly  with  a  magnitude  that  depends  mainly  upon 
the  thermocouple  wire  radius,  the  viscous  proper- 
ties of  the  sample  medium,  the  sound  intensity, 
and  the  frequency.    The  fast  rise  is  followed  by 
a  relatively  linear  rise  (in  the  absence  of 
thermal  conduction  processes)  that  is  a  result  of 
absorption  of  the  ultrasound  in  the  surrounding 
medium.    From  a  determination  of  the  slope  of  the 
1 inear  portion  of  the  thermoelectric  emf  response 
as  a  function  of  time,  the  absorption  coefficient 
can  be  calculated  using  the  following  relation, 


CnK 


p'-p 


21 


/dT\ 
^dt/n 


(5) 


where  a  is  the  absorption  coefficient  (cm"i),  p 
is  the  density  (g/cm^),  Cp  is  the  specific  heat  at 
constant  pressure  (cal/°C  g),  K  is  the  mechanical 
equivalent  of  heat  (4.18  J/cal),  I  is  the  acoustic 
intensity  (W/cm^),  and  (dT/dt)o  is  the  initial 
time  rate  of  change  of  temperature  due  to  absorp- 
tion in  the  medium  (°C/s).    Effects  limiting  the 


i 


pC  —  rubber  diaphragm 


position  of  probe 
for  determination 
of  absorbed  energy 


\ 


position  of  probe 
for  determination 
of  reflected  energy 
(A) 


castor  oil 
absorption 
chamber 


coupl ing 
medium 


position  of  excised 
lung  capsule 


sound 
tank 


Fig.  1, 


Schematic  diagram  of  the  experimental  arrangement  for  attenuation  and  velocity 
measurements  in  lung  using  the  standing  wave  method  [15]. 


45 


accuracy  of  the  method  include  any  changes  of  the 
thermal  properties  of  the  tissue  and/or  the  absorp- 
tion itself  with  temperature.    Provided  the  tem- 
perature rise  during  measurement  is  on  the  order 
of  one  degree  or  less  these  effects  will  be 
minimal.    Dunn  [18]  has  indicated  that  the  total 
uncertainty  in  the  determination  of  (dT/dt)n  is  of 
the  order  of  5  to  10  percent.  In  application  of  the 
transient  thermoelectric  method  the  initial  value 
of  a  is  determined  by  using  the  ultrasonic  in- 
tensity that  would  be  present  at  the  site  of  the 
junction  were  the  sample  absent.    The  measured 
depth  of  the  thermocouple  and  the  initial  value  of 
a  are  then  used  to  calculate  intensity  at  the  site 
of  the  junction  by  correcting  for  absorption  using 
an  iterative  method  [18]  until  the  value  of  o  con- 
verges (note  that  if  the  value  of  a  and/or  the 
depth  of  the  thermocouple  are  too  large,  the  value 
of  a  will  not  converge). 

This  technique  has  been  used  at  frequencies  as 
low  as  0.26  MHz  [19]  and  as  high  as  7  MHz  [20]  in 
tissues,  but  modifications  have  been  used  to  2  GHz 
in  fluid  media  [21].    The  lower  frequency  limit  is 
determined  by  the  value  of  the  absorption,  since 
too  low  an  a  yields  too  low  a  temperature  rise  to 
be  detected  accurately  without  having  to  increase 
the  exposing  intensity  to  unacceptable  levels. 
Also,  at  low  frequencies  and  absorptions  the  ini- 
tial viscous  heating  portion  of  the  transient 
thermal  emf  may  become  a  major  portion  of  the 
signal,  making  the  subtraction  process  a  relatively 
erroneous  one.    At  higher  frequencies  the  limita- 
tions on  accuracy  of  the  method,  as  described  for 
tissues,  are  imposed  by  (1)  availability  of  broad 
plane  wave  ultrasonic  fields,  (2)  the  size  of  the 
thermocouple  relative  to  the  wavelength  so  that 
the  field  is  affected  by  scattering  from  the  wire 
and  thermal  conduction  along  it,  and  ultimately 
by  (3)  nonconvergence  in  the  iterative  method  of 
depth  correction  due  to  high  absorption.    The  lat- 
ter two  limiting  effects  have  not  yet  been  trouble- 
some in  the  region  of  application.    The  method  has 
been  found  to  be  very  useful  and  easily  applied  to 
most  tissues  over  the  frequency  range  from  0.5  to 
4.0  MHz. 

4.    Absorption  and  Velocity  Measurement 
in  Biological  Liquids 

The  primary  mechanism(s)  for  the  absorption  of 
sound  in  biological  material  seems  to  occur  at 
the  macromolecul ar  level.    Carstensen,  Li,  and 
Schwan  [22]  discovered  that  the  acoustical  proper- 
ties of  blood  are  determined  largely  by  the 
protein  content,  and  that  the  absorption  coeffi- 
cient is  directly  proportional  to  the  protein 
concentration,  whether  in  solution  or  within  the 
cell.    Pauly  and  Schwan  [23]  have  demonstrated 
that  nearly  two-thirds  of  the  ultrasonic  absorp- 
tion of  beef  liver  lies  at  the  macromolecular 
level,  with  the  remaining  one-third  due  to  struc- 
tural features  of  the  tissue.    Since  the  primary 
mechanism  for  the  absorption  of  sound  in  tissue 
seems  to  occur  at  the  macromolecular  level,  it 
has  been  instructive  to  investigate  the  acoustic 
properties  of  simpler  systems  of  macromolecules, 
such  as  solutions  of  biopolymers,  with  the  hope 
that  such  studies  may  provide  details  applicable 
to  the  more  complex  systems. 

Three  measurement  systems  have  been  used  at  the 
Bioacoustics  Research  Laboratory  to  study  acoustic 
absorption  and  velocity  of  solutions  and  suspen- 


sions of  biomacromolecules  as  functions  of  mole- 
cular characteristics,  pH,  temperature,  concentra- 
tion, and  frequency  to  aid  in  the  elucidation  of 
the  mechanisms  of  interaction  between  ultrasound 
and  biological  tissues. 

A.    Absorption:    High  Frequency  Method 
(5  MHz  to  200  MHz) 

The  high  frequency  system  applies  pulsed  ultra- 
sound to  solutions  for  the  purpose  of  measuring 
absorption  and  velocity  [24-26].    Two  advantages 
of  pulsed  ultrasound  are  that  heating  effects  and 
standing  waves  are  virtually  eliminated.  Heating 
is  proportional  to  the  mean  power.    Using  a  10  ps 
pulse  at  a  repetition  frequency  of  300  pps  results 
in  a  signal  which  is  present  for  only  0.3  percent 
of  the  time.    Thus  the  average  intensity  is  much 
less  than  the  peak  intensity,  and  produces  a 
negligible  temperature  rise.    Since  the  signal  is 
on  for  such  a  small  fraction  of  the  time,  standing 
waves  are  not  produced.    The  basic  technique  was 
first  described  by  Pellam  and  Gait  in  1946  [27]. 
Two  transducers,  arranged  coaxial ly  face  each 
other  at  opposite  ends  of  a  cylindrical  tank.  The 
receiving  transducer  has  its  position  fixed  while 
the  sending  transducer  moves  toward  or  away  from 
it  at  a  constant  speed.    Assuming  the  relation 
governing  the  process  is 


p(x)  =  p(o)  e" 


(6) 


where  p(x)  is  the  wave  pressure  amplitude  at  the 
distance  x  from  the  transmitting  transducer,  p(o) 
is  the  pressure  amplitude  at  x  =  o  (i.e.,  at  the 
face  of  the  transmitting  transducer),  and  a  is  the 
absorption  coefficient,  the  natural  logarithm  of 
the  ultrasonic  pressure  amplitude  is  proportional 
to  the  signal  path  length.    The  proportionality 
constant,  i.e.,  the  absorption  coefficient,  can 
be  found  by  monitoring  the  pressure  as  a  function 
of  intertransducer  distance.    A  block  diagram  of 
the  system  is  shown  in  figure  2.    A  pulsed  rf 
signal  is  amplified,  passed  through  a  variable 


FREQUENCY 
SYNTHESIZER 


PULSE 

DOUBLE- 
BALANCED 
MIXER 

GENERATOR 

WIDE  BAND 
AMPLIFIER 


HATCHING 
NETWORK 


TRANSMITTING 
TRANSDUCER 


RECEIVING 
TRANSDUCER 


T 


PREAMPLIFIER 


SPECTRUM 
ANALYZER 


OPERATIONAL 
AMPLIFIER 


PULSE 

HEIGHT 

DETECTOR 

CHART 
RECORDER 

1  

ISOLATION 
AMPLIFIER 

DIGITAL 
VOLTMETER 


DIGITAL 
COMPUTER 


Fig.  2.    Block  diagram  of  instrumentation  used  for 
high  frequency  (5  to  200  MHz)  ultrasonic 
absorption  measurements  in  biological 
liquids. 


46 


attenuator,  and  impedance-matched  to  the  sending 
transducer.    Both  transducers  are  air  backed, 
gold-on-nickel  plated  X-cut  quartz  with  a  funda- 
mental resonant  frequency  of,  for  example,  3  MHz, 
and  operated  at  their  odd  harmonics.    The  wave 
passes  through  the  liquid,  where  it  is  attenuated, 
and  impinges  upon  the  receiving  transducer.  Some 
of  the  ultrasound  is  then  reconverted  into  an 
electric  signal,  while  the  remainder  is  reflected 
back  through  the  liquid.    A  pulse  height  monitor 
continually  detects  the  voltage  peak  of  the 
primary  signal  (which  is  linearly  proportional 
to  p(x))  as  the  transducers  are  moved  toward,  or 
away  from  each  other,  and  converts  it  to  its 
natural  logarithmic  value.    An  on-line  computer 
samples  the  output  of  the  pulse  height  detector 
(a  dc  voltage  proportional  to  the  pulse  height) 
via  a  digital  voltmeter,  calculates  the  attenua- 
tion coefficient,  and  corrects  for  diffraction 
effects  by  the  method  of  Del  Grosso  [28]  yielding 
the  true  absorption  coefficient.    Solute  absorp- 
tion is  obtained  by  subtracting  results  of  a  sol- 
vent measurement  from  a  solution  measurement. 
Approximately  one  liter  of  sample  is  required  for 
this  technique,  with  measurement  error  in  absorp- 
tion of  about  5  percent  over  the  approximate 
frequency  range  3  to  200  MHz.    Greater  errors  oc- 
cur in  low  absorbing  liquids  at  low  frequencies 
and  are  mainly  due  to  relatively  large  diffraction 
effects. 

B.    Velocity:    High  Frequency  Method 

The  acoustic  velocity  can  be  determined  in  the 
high  frequency  system  to  within  0.1  percent  by 
measuring  the  period  of  the  wave  in  the  liquid. 
This  is  accomplished  by  superimposing  a  reference 
signal  on  the  received  signal,  which  results  in 
an  interference  pattern.    If  the  acoustic  path 
length  is  changed  at  a  constant  rate,  Vp,  the 
period  of  this  pattern  is  T  =  (x/Vp)  where  X  is 
the  acoustic  wavelength.    Since  X  -  c/f,  where  f 
is  the  excitation  frequency,  the  acoustic  velocity 


positioning  rod 


is  given  as  c  =  TfVp.    The  period  T  is  measured 
using  a  time  interval  counter  to  determine  the 
time  required  for  the  interference  signal  to  go 
through  a  greater  number  (100)  of  maxima. 

C.    Absorption:    Low  Frequency  Method 
(0.3  MHz  to  20  MHz) 

In  the  frequency  range  of  0.3  to  20  MHz,  a 
pulse  technique  developed  by  Schwan  and  Carsten- 
sen  [29]  is  employed  in  order  to  avoid  having  to 
make  unreasonable  corrections  for  effects  of  dif- 
fraction phenomena.    A  sending  and  a  receiving 
transducer  face  each  other;  the  former  is  in  the 
reference  liquid  and  the  latter  in  the  test  liquid 
(see  fig.  3).    An  acoustic  window  separates  the 
two  liquids.    The  distance  between  the  trans- 
ducers remains  constant  while  the  entire  trans- 
ducer ensemble  is  moved  horizontally  at  a  constant 
speed  from  a  position  such  that  the  acoustic  path 
is  primarily  in  the  test  liquid  to  a  position  such 
that  it  is  primarily  in  the  reference  liquid.  The 
reference  liquid  is  chosen  to  be  dispersionless 
and  acoustically  well  characterized  with  a  sound 
velocity  as  near  to  that  of  the  unknown  as  pos- 
sible.   Water  is  an  appropriate  choice  for  dilute 
aqueous  solutions.    Since  the  path  length  is 
constant,  diffraction  effects,  due  to  the  in- 
equality of  the  velocities  in  the  two  liquids, 
are  corrected  by  a  computer  using  the  method  of 
Del  Grosso  [28].    Liquid  volumes  of  1  to  4  liters 
are  used  in  each  chamber,  depending  upon  frequency. 
As  with  the  high  frequency  system,  measurement 
error  ranges  from  2  to  10  percent.    The  pressure 
amplitude  in  the  receiving  transducer  is  given  by 


-aw(d- 


(7) 


where  p^  is  the  pressure  amplitude  at  the  sending 
transducer,  and  the  other  symbols  are  described 
by  figure  3.    This  equation  can  be  rewritten  as 


1 n  p„  =  In  Pt 


o,  d  + 
w 


(8) 


transducer 


acoustic  window 
 X  


transducer 


Figure  3.    Schematic  diagram  of  apparatus  used  for  ultrasonic  absorption  and  velocity  measurements 
at  low  frequencies  (0.3  MHz  to  20  MHz)  in  biological  liquids  [24]. 


47 


and  can  be  obtained  graphically  from  the  slope 
since       is  known. 

As  the  acoustic  window  (polyethylene)  reflects 
a  negligible,  but  constant  amount  of  the  acoustic 
energy,  it  will  affect  the  results  only  by  chang- 
ing the  intercept,  i.e.,  the  slope  is  not  affect- 
ed.   Signal  processing  to  obtain  the  desired 
absorption  coefficient  closely  resembles  that 
used  in  the  high  frequency  system,  described 
above  [14,24]. 


nally  developed  by  Eggers  [31],  which  employs  two 
X-cut  quartz  disks,  with  a  fundamental  frequency 
of  the  order  of  2  MHz,  separated  by  a  lucite  ring, 
and  forming  a  cylindrical  cavity  in  which  the 
specimen  solution  is  placed.    One  transducer 
serves  as  a  transmitter,  and  the  other  a  receiver. 
When  the  transmitting  transducer  is  excited  with  a 
sinusoidal  voltage,  standing  waves  in  the  specimen 
solution  result  at  particular  frequencies  that 
obey  the  resonance  equation 


D.    Velocity:    Low  Frequency  Method 

Velocity  of  the  acoustic  wave  can  be  determined 
with  the  low  frequency  system  to  0.01  percent 
[30]  by  superimposing  the  received  and  reference 
signals  in  the  same  fashion  as  was  done  for  the 
high  frequency  system.    If  the  velocities  in  the 
test  and  reference  liquids  are  different,  moving 
the  transducer  assembly  will  change  the  acoustic 
path  length.    By  observing  the  interference  pat- 
tern on  the  oscilloscope,  one  can  position  the 
transducers  such  that  a  maximum  occurs.    If  this 
is  position  1  in  figure  3,  the  number  n,  of 
acoustic  wave  lengths  between  the  transducers  is 
given  by 

n=^^  +  ^  (9) 


If  the  transducers  are  moved  a  distance.  Ax,  such 
that  the  interference  pattern  undergoes  an  inte- 
gral number,  m,  of  2t\  phase  shifts,  the  above  ex- 
pression becomes 


d  -  X  +  Ax  ,  x 

n  ±  m  =  +  - 


Ax 


(10) 


where  the  positive  sign  applies  when  the  velocity 
in  the  test  liquid  is  larger  than  that  of  the  re- 
ference liquid,  as  is  usually  the  case.  Eliminat- 
ing n  between  these  equations,  expressing  wave- 
lengths in  terms  of  velocity  and  frequency,  and  a 
rearranging  of  terms  yields 


Pq^q 


Cntan 


PcC 


s^s 


r-tan  1 
[cotanj 


2  f. 


(12) 


where  pq  and  c„  are  the  density  and  speed  of  ' 
sound  in  the  quartz  transducers,  and  Pg  and  c^ 
are  the  corresponding  quantities  in  the  solution. 
The  fundamental  frequency  of  the  transducer  is 
fg,  and  of  the  liquid-filled  cavity,  f^.    At  the 
resonant  frequencies  fp,  the  receiving  trans- 
ducer registers  pronounced  voltage  peaks,  whose 
frequency  separation  depends  on  the  sound  velo- 
city of  the  specimen  solution,  and  the  half- 
power  (3  dB)  bandwidth  Af  of  which  is  related  to 
the  attenuation  per  wavelength,  aX.  Eggers 
[31,32]  has  shown  that  the  velocity  of  sound  Vf 
in  an  unknown  medium  may  be  related  to  the  velo- 
city of  sound  in  some  reference  medium  v,,  by  the 
expression 


v_ 


D 


1  +  2(DfZf  -  D^Ir) 


^q^q 


(13) 


where  If  and  Z„  are  respectively,  the  acoustic 
impedance  of  tne  unknown  and  reference  medium, 
and       and  Dp  are  the  respective  separations  in 
frequency  units  between  adjacent  resonances  for 
the  liquid  and  reference  media,  and  f^  and  I„ 
respectively,  are  the  frequency  of  the  quartz  and 
its  acoustic  impedance.    Equation  13  may  be  ap- 
proximated by  a  simpler  expression  (with  a  dif- 
ference of  only  a  few  parts  per  thousand)  as 


Cv  = 


(11) 


fAx 


Note  that  some  other  procedure  must  be  performed 
to  determine  the  correct  sign. 

E.    Absorption  and  Velocity  Measurements 
in  Small  Liquid  Volumes: 
Resonant  Cavity  Method 

The  various  methods  for  determining  the  acoustic 
propagation  properties  of  liquids  discussed  thus 
far  were  devised  with  little  consideration  for  the 
volume  of  material  necessary  for  measurement.  The 
minimum  volume  required  in  the  high  frequency 
system  is  500  ml,  with  greater  volumes  generally 
required  at  lower  frequencies  to  avoid  effects  due 
to  diffraction  phenomena.    The  necessity  for  having 
such  large  volumes  available  becomes  a  serious 
problem  when  biological  macromolecules  are  to  be 
treated  in  solutions  (usually  in  concentrations  of 
10  percent  by  weight),  since  only  a  few  can  be 
examined  within  the  bounds  of  reasonable  economics. 
It  is  possible  to  reduce  the  specimen  volume  to 
less  than  50  ml  using  a  resonance  technique  or^'^i- 


&Cf 


6f, 


(14) 


where  SCf  is  the  velocity  difference  between  the 
unknown  and  reference  media,  Sf^  is  the  difference 
between  corresponding  resonances,  and  Cf  is  the 
velocity  of  the  reference  medium  at  a  frequency 
fp.    In  general,  velocity  measurements  are  more 
difficult  to  perform  than  absorption  measurements 
due  to  temperature  drift  and  other  instabilities 
of  the  unit.    The  attenuation  of  the  specimen 
solution  may  be  calculated  from  the  examination 
of  the  quality  factor  Q  of  the  liquid  filled 
cavity,  defined  as  the  frequency  fp  divided  by  the 
3  dB  bandwidth,  Afp,  of  the  resonance.    This  quali 
ty  factor  is  a  function  of  the  mechanical  clamping 
and  the  attenuation  per  wavelength  aX  of  the  ultra 
sonic  energy  and  is  expressed  as. 


Q  = 


Afn 


oX 


(15) 


The  measured  Q,  however,  includes  losses  associat- 
ed with  attenuation  in  the  solvent,  as  well  as ^ 
those  arising  from  diffraction,  wall  effects,  im- 
perfect reflections  at  the  quartz  surface,  etc. , 


48 


in  addition  to  the  desired  excess  attenuation  due 
to  the  solute.    Assuming  all  of  these  energy  losses 
are  additive  [32],  the  measured  quality  factor  Q 
is  given  by 


%eas     ^solute  ^extra 

where  Qsolute  is  the  quality  factor  due  to  sound 
absorption  in  the  solute,  and  Qextra  includes  sol- 
vent and  other  cell  losses  previously  discussed. 
The  excess  solute  absorption  can  be  obtained  by 
means  of  a  reference  measurement  in  the  same  cell 
at  the  same  frequencies  with  the  reference  liquid, 
having  equal  or  very  similar  sound  velocity  to  in- 
sure the  same  sound  field  pattern  for  both  mea- 
surements.   The  excess  absorption  per  wavelength 
in  the  specimen  solution  is  then  obtained  from 

-lexcess        \     ^n  / 

where  Af^  and  Af^  are  the  corresponding  3  dB 
bandwidtns  of  the  n^h  resonant  peak  in  the  sample 
and  reference  liquid,  respectively.    The  applica- 
tion of  an  overpressure  (usually  about  10  psi) 
causes  a  slight  concavity  of  the  transducers, 
which  reduces  diffraction  and  boundary  effects, 
and  thereby  reduces  the  minimum  frequency  of 
measurement  [33].    This  technique  has  been  used 
in  our  laboratory  over  the  frequency  range  from 
0.5  to  10  MHz. 

5.    Absorption  and  Velocity  Measurements 
of  Shear  Waves  in  Biological  Specimens 

A  system  based  on  a  pulse  superposition  tech- 
nique allows  measurement  of  the  shear  acoustical 
properties  of  biological  materials  of  interest 
[34].    The  technique  involves  phase-amplitude 
balance  to  measure  the  magnitude  and  phase  angle 
of  the  reflection  coefficient  of  a  shear  wave 
impinging  on  a  quartz-sample  interface,  and  using 
the  known  impedance  of  the  quartz,  the  complex 
shear  specific  impedance  of  the  sample  being  in- 
vestigated is  obtained  [35].    From  the  shear  im- 
pedance and  density,  it  is  possible  to  calculate 
the  dynamic  shear  stiffness  pi,  dynamic  shear 
viscosity  \i2'  shear  velocity  cs,  and  shear  absorp- 
tion coefficient  ag  as  shown  in  the  following 
relationships: 


impedance  and  Xs  is  the  imaginary  part  of  the 
specific  shear  acoustic  impedance.    This  method 
has  been  used  to  measure  the  shear  acoustic  prop- 
erties of  tissues  and  other  biological  specimens 
[36-38].    An  improved  system  utilizing  a  gated 
carrier  as  opposed  to  a  pulsed  oscillator  approach 
and  an  improved  ultrasonic  unit  of  beveled  AT 
quartz  is  being  developed  that  will  be  useful 
over  the  frequency  range  from  2  to  greater  than 
20  MHz  and  for  temperatures  from  10  to  40  °C. 
These  and  additional  improvements  should  provide 
accuracies  for  the  real  and  imaginary  parts  of  the 
specific  acoustic  impedance  v/ithin  ±  100  mech-ohm/cm^ 
at  10  MHz,  allowing  more  accurate  measurement  of  im- 
pedance close  to  that  for  water. 

6.    Concluding  Remarks 

The  measuring  techniques  for  ultrasonic  absorp- 
tion, attenuation  and  velocity  in  biological  speci- 
mens discussed  above  have  been  only  those  utilized 
at  the  Bioacoustics  Research  Laboratory.  There 
are,  however,  techniques  employed  elsewhere  which 
may  provide  the  same  information  in  a  more  effi- 
cient manner.    For  example,  the  above  discussed 
methods  provide  information  at  discrete  frequen- 
cies and  loci,  or,  for  spatial  averages.  Spec- 
trum analysis  methods  [39-41]  have  been  promoted 
to  permit  measurements  over  a  somewhat  broadened 
frequency  range.    The  measurement  of  the  spatial 
distribution  of  ultrasonic  properties  such  as  at- 
tenuation or  velocity  within  a  specimen  may  be 
possible  using  the  algebraic  reconstruction  from 
two-dimensional  acoustic  projections  [42,43],  al- 
though only  velocity  reconstruction  has  been  con- 
sidered practical  so  far. 

Recent  studies  have  also  considered  improve- 
ments regarding  possible  error  in  established 
methods  of  ultrasonic  parameter  measurements. 
Artifacts  in  acoustic  attenuation  due  to  phase 
cancellation  effects  have  been  described  by 
Marcus  and  Carstensen  [6],  who  compared  a  piezo- 
electric receiver  (phase-sensitive)  with  a 
radiation-force  receiver  (phase-insensitive),  and 
also  by  Busse  et  al .   [44],  comparing  the  piezo- 
electric receiver  with  an  acousto-electric  re- 
ceiver.   Both  studies  found  that  measurements 
using  piezoelectric  (phase-preserving)  receivers 
yielded  higher  apparent  attenuation  values  than 
those  obtained  using  phase-insensitive  receivers. 
Phase  cancellation  artifacts  are  thought  to  be 
the  source  of  this  error. 


Pi  =  M_Lis  (18) 

P  The  authors  acknowledge  gratefully  the  partial 

support  for  portions  of  the  activities  described 
Y  herein  by  grants  from  the  National  Institutes  of 

V2  =  (19)  Health. 


2  Acknowledgment 


up 


Rc  + 


pojX 


References 


s     '^s  (20)  [1]    Rooney,  J.  A.,  Determination  of  acoustic 

pR  power  outputs  in  the  microwatt-milliwatt 

range.  Ultrasound  Med.  Biol.  1,  13-16  (1973). 


^—  ^^^^  [2]    O'Brien,  W.  D. ,  Jr.,  Ultrasonic  Dosimetry, 

in  Ultrasound:    Its  Application  in  Medicine 


where  p  is  the  density,  u  is  the  angular  frequency, 
Rj  is  the  real  part  of  the  specific  shear  acoustic 


and  Biology,  F.  J.  Fry,  ed. ,  Chap. 
(Elsevier,  Amsterdam,  in  press] 


49 


[3]    Kossoff,  G.,  Balance  technique  for  the 
measurement  of  very  low  ultrasonic  power 
outputs,  J.  Acoust.  Soc.  Am.  38^,  880-881 
(1965). 

[4]    Hill,  C.  R. ,  Calibration  of  ultrasonic  beams 
for  biomedical  applications,  Phys.  Med. 
Biol .  15,  241-248  (1970). 


[5]    Pohlman,  R. ,  Uber  die  Absorption  des  Ultra- 
schalls  im  menschlichen  Gewebe  und  ihre 
Abhangigkeit  von  der  Frequenz,  Physik  Z.  40, 

159-  161  (1939). 

[6]    Marcus,  P.  W.  and  Carstensen,  E.  L.,  Prob- 
lems with  absorption  measurements  of  in- 
homogeneous  solids,  J.  Acoust.  Soc.  Am.  58 , 
1334-1335  (1975). 

[7]    Dussik,  K.  T.,  Fritch,  D.  J.,  Kyriazidou,  M.  , 
and  Sear,  R.  S.,  Measurements  of  articular 
tissue  with  ultrasound,  J.  Phys.  Med.  37 , 

160-  165  (1958). 

[8]    Bauld,  T.  J.  and  Schwan,  H.  P.,  Attenuation 
and  reflection  of  ultrasound  in  canine  lung 
tissue,  J.  Acoust.  Soc.  Am.  56,  1630-1637 
(1974). 

[9]    Wladimiroff,  J.  W.,  Craft,  I.  L.,  and  Talbert, 
D.  G. ,  In  vitro  measurement  of  sound  velocity 
in  human  fetal  brain,  Ul trasound  Med.  Biol . 
I,  377-382  (1975). 

[10]    Buschmann,  W. ,  Voss,  M.,  and  Kemmerling,  S. , 
Acoustic  properties  of  normal  human  orbit 
tissues,  Ophtal.  Res.  I,  354-364  (1970). 

[11]    Lange,  J.  N.,  Group  velocity  dispersion  due 
to  pulse  reflection  from  a  frequency  depend- 
ent boundary  impedance,  J.  Acoust.  Soc.  Am. 
40,  998-1001  (1966). 

[12]    Goldman,  D.  E.  and  Richards,  J.  R.,  Measure- 
ment of  high  frequency  sound  velocity  in 
mammalian  soft  tissues,  J.  Acoust.  Soc.  Am. 
26^,  981-983  (1954). 

[13]    Hueter,  T.  F.  ,  Viscoelastic  Losses  in  Tis- 
sues in  the  Ultrasonic  Range,  Wright  Air 
Development  Center  Technical  Report  57-706 
(1958). 

[14]    Fry,  W.  J.  and  Dunn,  F. ,  Ultrasound: 
Analysis  and  Experimental  Methods  in 
Biological  Research,  in  Physical  Techniques 
in  Biological  Research,  W.  L.  Nastuk,  ed.. 
Vol.  IV,  Chap.  6,  pp.  261-394  (Academic 
Press,  New  York,  1962). 

[15]    Dunn,  F.  and  Fry,  W.  J.,  Ultrasonic  absorp- 
tion and  reflection  by  lung  tissue,  Phys. 
Med.  Biol.  5^,  401-410  (1961). 

[16]    Fry,  W.  J.  and  Fry,  R.  B. ,  Determination 
of  absolute  sound  levels  and  acoustic  ab- 
sorption coefficients  by  thermocouple 
probes--theory ,  J.  Acoust.  Soc.  Am.  26, 
294-310  (1954).  ^ 

[17]    Fry,  W.  J.  and  Fry,  R.  B.,  Determination 
of  absolute  sound  levels  and  acoustic  ab- 


sorption coefficients  by  thermocouple 
probes--experiment,  J.  Acoust.  Soc.  Am.  26, 
311-317  (1954). 

[18]    Dunn,  F.,  Temperature  and  amplitude  depend- 
ence of  acoustic  absorption  in  tissue,  J_. 
Acoust.  Soc.  Am.  34,  1545-1547  (1962). 

[19]    Dunn,  F.  and  Brady,  J.  K.,  Temperature  and 
Frequency  Dependence  of  Ultrasonic  Absorp- 
tion in  Tissue,  in  Proceedings  8th  Inter- 
national Congress  on  Acoustics,  Vol.  1, 
p.  366c  (Goldcrest  Press,  Trowbridge , 
Wilts.,  1974). 

[20]    Brady,  J.  K.,  Goss,  S.  A.,  Johnston,  R.  L., 
O'Brien,  W.  D.,  Jr.,  and  Dunn,  F.,  Ultra- 
sonic propagation  properties  of  mammalian 
testes,  J.  Acoust.  Soc.  Am.  60_,  1407-1409 
(1976). 

[21]    Dunn,  F.  and  Breyer,  J.  E.,  Generation  and 
detection  of  ultra-high-frequency  sound  in 
liquids,  J.  Acoust.  Soc.  Am.  34,  775-778 
(1962). 

[22]    Carstensen,  E.  L.,  Li,  K. ,  and  Schwan,  H.  P., 
Determination  of  the  acoustic  properties  of 
blood  and  its  components,  J.  Acoust.  Soc. 
Am.  25,  286-289  (1953). 

[23]  Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of 
absorption  of  ultrasound  in  liver  tissue, 
J.  Acoust.  Soc.  Am.  50,  692-699  (1971). 

[24]    O'Brien,  W.  D. ,  Jr.,  The  Absorption  of 

Ultrasound  in  Aqueous  Solutions  of  Biologi- 
cal Polymers,  Ph.D.  Thesis,  University  of 
Illinois,  Urbana,  Illinois  (1970). 

[25]    Goss,  S.  A.,  The  Concentration  Dependence 
of  the  Ultrasonic  Absorption  in  Aqueous 
Solutions  of  Bovine  Serum  Albumin,  M.S. 
Thesis,  University  of  Illinois,  Urbana, 
Illinois  (1974). 

[26]    Kessler,  L.  W.,  Hawley,  S.  A.,  and  Dunn,  F. , 
Semi-automatic  determination  of  ultrasonic 
velocity  and  absorption  in  liquids, 
Acustica  24,  105-107  (1971). 

[27]    Pellam,  J.  R.  and  Gait,  J.  K.,  Ultrasonic 

propagation  in  liquids:    I.    Application  of 
pulse  technique  to  velocity  and  absorption 
measurements  at  15  megacycles,  J.  Chem. 
Phys .  14'  608-614  (1946). 

[28]    Del  Grosso,  V.A.,  Systematic  Errors  in  Ultra- 
sonic Propagation  Parameter  Measurements 
Part  I--Effect  of  Free-Field  Diffraction, 
U.S.  Naval  Research  Laboratory  Report  6026 
(1964). 

[29]    Schwan,  H.  D.  and  Carstensen,  E.  L.,  Ultra- 
sonics aids  diathermy  experiments.  Elec- 
tronics 25,  216-220  (1952). 


[30]    Carstensen,  E.  L.,  Measurement  of  dispersion 
of  velocity  of  sound  in  liquids,  J.  Acoust. 
Soc.  Am.  26,  858-861  (1954). 

[31]    Eggers,  F.,  Eine  Resonatormethode  zur  Be- 


50 


stimmung  von  Schal 1 --Geschwindigkei t  und 
Dampfung  an  gen'ngen  Fliissigkeitsmengen, 
Acustica  19.  323-329  (1968). 

[32]    Eggers,  F.  and  Funck,  Th.,  Ultrasonic  mea- 
surements with  milliliter  liquid  samples  in 
the  0.5-100  MHz  range,  Rev.  Sci.  Instrum. 
44,  969-977  (1973). 

[33]    Eggers,  F.,  Funck,  Th.,  and  Richman,  K.  H., 
High  Q  ultrasonic  liquid  resonators  with 
concave  transducers.  Rev.  Sci.  Instrum.  47, 
361-367  (1976). 

[34]    Frizzell,  L.  A.,  Carstensen,  E.  L.,  and 

Dyro,  J.  F.  ,  Shear  properties  of  mammalian 
tissues  at  low  megahertz  frequencies,  J^. 
Acoust.  Soc.  Am.  60,  1409-1411  (1976). 

[35]    Moore,  R.  S.  and  McSkimin,  H.  J.,  Dynamic 

Shear  Properties  of  Solvents  and  Polystyrene 
Solutions  from  20  to  300  MHz,  in  Physical 
Acoustics,  W.  P.  Mason  and  R.  N.  Thurston, 
eds..  Vol.  VI,  Chap.  4,  pp.  167-242  (Academic 
Press,  Inc.,  New  York,  1970). 

[36]    Dyro,  J.  F.,  Ultrasonic  Study  of  Material 
Related  to  Atherosclerotic  Plaque--Dynamic 
Viscoelastic  Properties  of  Cholestoric 
Esters,  Ph.D.  Thesis,  University  of  Penn- 
sylvania, Philadelphia,  Pennsylvania  (1972). 

[37]    Dyro,  J.  F.  and  Edmonds,  P.  D.,  Ultrasonic 
absorption  and  dispersion  in  cholesteryl 
esters,  Mol.  Cryst.  Liq.  Cryst.  25,  175-193 
(1974). 

[38]    Frizzell,  L.  A.,  Ultrasonic  Heating  of 

Tissues,  Ph.D.  Thesis,  University  of  Rochester, 
New  York  (1975). 


[39]-   Chivers,  R.  C.  and  Hill,  C.  R.,  Ultrasonic 
attenuation  in  human  tissue.  Ultrasound  Med. 
Biol.  2,  25-29  (1975). 


[40]    Lizzi,  F.,  Katz,  L.,  St.  Louis,  L.,  and 
Coleman,  D.  J.,  Applications  of  Spectral 
Analysis  in  Medical  Ultrasonography,  Ultra- 
sonics 14,  77-80  (1976). 

[41]    Le  Croissette,  D.  H.  and  Heyser,  R.  C, 
Attenuation  and  Velocity  Measurements  in 
Tissue  Using  Time  Delay  Spectrometry,  in 
Ultrasonic  Tissue  Characterization,  M. 
Linzer,  ed..  National  Bureau  of  Standards 
Spec.  Publ.  453,  pp.  81-95  (U.S.  Government 
Printing  Office,  Washington,  D.C.,  1976). 

[42]    Greenleaf,  J.  F.,  Johnson,  S.  A.,  Lee,  S.  L. 
Herman,  G.  T.,  and  Wood,  E.  H.,  Algebraic 
Reconstruction  of  Spatial  Distributions  of 
Acoustic  Absorption  Within  Tissue  From  Their 
Two-Dimensional  Acoustic  Projections,  in 
Acoustical  Holography,  P.  S.  Green,  ed.. 
Vol .  5,  pp.  591-603  (Plenum  Press,  New  York, 
1974). 

[43]    Greenleaf,  J.  P.,  Johnson,  S.  A.,  Samayoa, 

W.  F.,  and  Duck,  F.  A.,  Algebraic  Reconstruc 
tion  of  Spatial  Distributions  of  Acoustic 
Velocities  in  Tissue  From  Their  Time  of 
Flight  Profiles,  in  Acoustical  Holography, 
N.  Booth,  ed..  Vol.  6,  pp.  71-90  (Plenum 
Press,  New  York,  1975). 

[44]    Busse,  L.  J.,  Miller,  J.  G.,  Yuhas,  D.  E., 
Mimbs,  J.  W.,  Weiss,  A.  N.,  and  Sobel,  B.  E. 
Phase  Cancellation  Effects:    A  Source  of  At- 
tenuation Artifact  Eliminated  by  a  CdS 
Acoustoelectric  Receiver,  in  Ultrasound  in 
Medicine,  D.  White,  Ed.,  Vol.  3,  pp.  1519- 
1535  (Plenum  Press,  New  York,  1977). 


51 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


A  DEVICE  FOR  MEASURING  ULTRASONIC  PROPAGATION  VELOCITY  IN  TISSUE 


Bruce  D.  Sollish 

Harry  De  Jur  Diagnostic  Instrumentation  Laboratory 
Department  of  Electronics 
Weizmann  Institute  of  Science 
Rehovot,  Israel 


This  paper  discusses  a  device  capable  of  measuring  propagation  velocity  in  tissue, 
both  excised  and  in  vivo.    The  device  produces  a  4-digit  decimal  readout  of  propagation 
velocity  in  the  specimen  relative  to  that  in  water.    As  presently  constituted  the  de- 
vice is  capable  of  1  velocity  measurement  per  second,  with  one  additional  second  re- 
quired for  printout  of  the  result. 

The  theory  of  the  device  is  described,  and  experimental  results  are  presented  for 
solids  and  soft  tissues. 

Key  words:    Propagation  velocity;  reflection  technique;  solids;  tissue;  ultrasound. 


1.  Introduction 

Ultrasonic  propagation  velocity  has  been  mea- 
sured in  a  variety  of  soft  tissues.    In  a  review 
article,  Wells  [iy  has  compiled  some  of  the  more 
significant  velocity  data  published  in  the  litera- 
ture.   It  would  appear  that  accurate  measurement 
of  velocity  could  help  in  characterizing  soft 
tissues.    Kossof f  et  al .  [2],  for  example,  have 
measured  ultrasonic  velocity  in  the  human  female 
breast  and  have  concluded  that  the  results  as  to 
characteristics  of  tissue  type  correlate  reason- 
ably well  with  the  findings  of  mammography. 

One  problem  in  velocity  measurement  is  the 
need  for  determining  thickness  of  the  tissue 
along  the  path  of  propagation.    Thickness  can  be 
measured  directly  or  by  the  method  of  equivalent 
water  path  given  by  Kossoff  [2,3]  for  transmis- 
sion and  reflection  velocity  measurement  tech- 
niques.   For  each  new  propagation  path  through 
the  tissue,  another  thickness  measurement  must 
be  made.    The  process  of  measuring  velocity  in 
a  variety  of  paths  through  the  tissue  could 
therefore  be  quite  time  consuming. 

In  order  to  eliminate  the  need  to  determine 
tissue  thickness  along  the  propagation  path,  a 
new  pulse-echo  method  for  measuring  ultrasound 
velocity  has  been  developed.    This  technique 
utilizes  a  reflecting  surface  at  a  fixed  dis- 
tance from  the  transducer.    The  overall  propaga- 
tion path  consists  of  a  suitable  medium,  such 
as  water,  and  the  tissue  sample. 

Since  only  the  overall  distance  between  the 
transducer  and  reflector  need  be  constant,  the 
thickness  of  the  tissue  need  not  be  measured. 
The  technique  can  therefore  be  readily  applied 
in  velocity  measurements  in  different  samples  or 
through  different  paths  in  the  same  sample. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


without  repeated  thickness  calculations.  Indeed, 
the  velocity  data  can  be  used  to  compute  thick- 
ness in  the  specimen  along  the  propagation  path, 
which  can  be  compared  with  the  thickness  as 
measured  directly,  for  checking  the  accuracy  of 
the  measurement. 

A  prototype  of  a  device  incorporating  this 
reflection  technique  has  been  built,  for  mea- 
surements of  propagation  velocity  in  solids  and 
in  tissue.    The  following  sections  describe  in 
more  detail  the  reflection  technique,  the  proto- 
type device  implementing  the  technique,  and  re- 
sults of  velocity  measurements  in  solids  and  in 
tissue. 

2.    Description  of  the  Technique 

The  method  used  for  measuring  propagation 
velocity  is  illustrated  in  figure  1.    A  flat 
reference  reflecting  surface  is  positioned  at  a 
convenient  fixed  distance  from  the  ultrasonic 
transducer.    Both  are  immersed  in  a  suitable 
propagating  medium  such  as  distilled  water.  An 
A-mode  pulse-echo  device  displays  an  echo  due  to 
the  reflecting  surface.    This  echo  is  a  reference 
echo  for  the  remainder  of  the  velocity  measure- 
ment. 

The  tissue  specimen  is  next  interposed  between 
the  transducer  and  the  reflector  as  shown  in  the 
figure.    The  A-scope  is  adjusted  to  display  the 
echoes  corresponding  to  the  following  interfaces: 
the  anterior  boundary  of  the  specimen,  the  pos- 
terior boundary  of  the  specimen,  and  the  re- 
flector surface  as  returned  through  the  specimen. 
The  reflector  echo  is  shifted  in  time  from  its 
reference  position  if  the  velocity  in  the  speci- 
men is  different  from  that  in  the  medium. 

As  shown  in  the  figure,  there  are  in  general 
two  configurations  for  velocity  measurement.  In 
the  first  configuration,  the  specimen  is  posi- 
tioned somewhere  in  between  the  transducer  and 
the  reflector.     In  this  case  four  echoes  of  in- 


53 


transducer 


water 


reference  reflector 


initial  pulse 


transducer 


reference  echo 


initio!  pulse 


transducer 


reference  reflector 


onginol  reference 
echo 

displaced  reference 
echo 


target  reference  reflector 
n-t,  J 


Tllb 


reference  echo 


initial  pulse 


anterior  posterior  /  displaced 
target  target  /  reference 
echo        echo   /  echo 


Fig.  1.    Reference  reflector  technique  for  velocity 
measurements . 

a)  initial  calibration 

b)  target  interposed  between  transducer 
and  reference  reflector 

c)  target  in  contact  with  reference 
ref 1 ector . 

terest  are  produced.    In  the  second  configuration, 
the  specimen  is  positioned  so  that  its  posterior 
boundary  is  coincident  with  the  reflector.  In 
this  case,  there  are  only  three  echoes  to  con- 
sider.   The  second  configuration  is  readily 
adapted  for  use  in  velocity  measurements  made  via 
a  water  bag,  in  which  case  the  specimen  and  the 
reflector  lie  outside  the  water  bag. 

For  either  configuration,  velocity  in  the 
sample  is  calculated  as  follows.    Denote  the  dis- 
tance between  the  transducer  and  reflector  by  L, 
the  thickness  of  the  specimen  by  i,  the  velocity 
in  the  medium  by  Vq,  and  the  velocity  in  the 
specimen  by  V.    Let  the  time  between  the  anterior 
and  posterior  echoes  from  the  specimen  be  T^,  and 
let  the  time  displacement  of  the  reflector  echo 
due  to  interposition  of  the  specimen  be  T2.  Then 
the  following  equations  apply  for  Tj  and  T2: 


Ti  =  2e/V 


-  2L 

Vo 


T,  =  ^ 


[20: 

L  Vn 


21" 
V 


=  2?  i 


I 

Vo 


(1) 


(2) 


From  these  equations,  the  ratio  of  propagation 
velocity  in  the  specimen  to  that  in  the  medium  is 


V/Vo  =  1  +  T2/T1 


(3) 


We  note  that  To  is  positive  if  the  reflector  echo 
is  displaced  toward  the  transducer  when  the  speci- 
men is  in  position,  and  negative  if  the  reflector 
is  displaced  away  from  the  transducer. 

The  thickness  of  the  specimen  along  the  path 
of  propagation  can  be  calculated  from  the  above 
equations,  as 


where 


5./f.o  =  V/Vr 


VoTi 


2 


(4) 
.  (5) 


We  note  that  Sq  is  the  thickness  of  the  specimen 
measured  by  the  A-scope  calibrated  for  velocity 
in  the  medium. 

The  significance  of  eqs.  (1)  through  (5)  is 
that  the  ratio  of  T2  to  T^  is  independent  of  the 
thickness  i  of  the  specimen.    Accurate  measure- 
ment of  times  Tj  and  T2  will  give  an  accurate 
measurement  of  velocity  in  the  specimen  relative 
to  velocity  in  the  medium.    Even  if  the  specimen 
is  repositioned  or  if  a  different  specimen  is 
examined,  Tj  and  T2  will  both  change  accordingly, 
as  long  as  the  transducer  to  reflector  distance 
remains  unchanged.    Thus,  the  technique  enables 
rapid  velocity  measurements  in  a  variety  of  tis- 
sues, as  long  as  the  reference  echo  is  returned 
through  the  specimen. 

3.    Description  of  the  Device 

A  block  diagram  of  a  device  utilizing  the  re- 
ference reflector  technique  for  measuring  ultra- 
sound velocity  is  shown  in  figure  2.    An  A-scope 
and  an  oscilloscope  are  required.    In  addition, 
a  printer  may  be  added  for  hard  copy  output  of 
velocity  data. 


sync  out  \ 


a-scope 


tinning 
circuitry 

program 
circuitry 

data 

acquisition 
circuitry 

calculator 
circuitry 

digital 
display 


transducer 


auxiliary 
display 
oscilloscope 


Fig.  2.    Block  diagram  of  velocity  measurement 
device  and  associated  equipment. 


Operation  of  the  system  is  as  follows.  Before 
the  specimen  is  immersed  in  water,  the  A-scope  is 
adjusted  to  display  the  echo  due  to  the  reference 
reflector.  The  A-mode  ("video")  signal  is  fed  to 
the  data  acquisition  circuit,  passing  through  a 


54 


time-gated  threshold  detector  and  a  pulse  shaper. 
The  processed  echo  from  the  reflector  is  dis- 
played on  the  oscilloscope,  together  with  a  pulse 
whose  position  on  the  display  is  variable.  The 
user  presets  the  time  delay  of  the  variable  pulse, 
so  that  the  pulse  overlaps  the  reference  echo,  as 
seen  on  the  oscilloscope  and  confirmed  by  a  front- 
panel  LED  indicator.    This  procedure,  which 
stores  in  the  data  acquisition  circuitry  the  prop- 
agation time  from  the  transducer  to  the  reflector, 
is  performed  only  once  before  the  specimen  is  in- 
serted.   At  any  time,  calibration  may  be  checked 
by  removing  the  specimen,  and  noting  if  the  over- 
lap condition  is  resumed. 

Following  storage  of  the  reference  echo,  the 
specimen  is  interposed  between  the  transducer  and 
the  reflector.    The  data  acquisition  circuit,  by 
adjustment  of  the  time-gated  threshold  detector, 
eliminates  internal  echoes  from  within  the  sample. 
The  present  velocity  device  implements  the  con- 
figuration of  figure  Ic.    Therefore  only  two 
echoes  remain  after  processing:    the  anterior 
echo  of  the  specimen  and  the  echo  at  the  inter- 
face between  the  specimen  and  reflector. 

The  data  acquisition  circuit  now  contains 
three  pulses:    the  two  echoes  processed  from  the 
tissue  specimen  and  the  preset  pulse  supplied  by 
the  user  as  described  earlier.    Two  gates  are  de- 
rived from  these  pulses.    The  first  gate  is 
opened  by  the  anterior  pulse  and  closed  by  the 
specimen-reflector  boundary  echo.    The  second 
gate  is  opened  by  the  specimen-reflector  boundary 
echo  and  closed  by  the  preset  pulse  or  opened  by 
the  preset  pulse  and  closed  by  the  specimen- 
boundary  echo,  depending  on  which  signal  arrives 
first.    In  the  latter  case,  a  flag  is  sent  out 
to  the  calculator  circuit  to  perform  a  subtrac- 
tion instead  of  an  addition,  as  discussed  later. 

Each  gate  enables  one  of  two  4-digit  BCD 
counters  clocked  at  a  rate  of  15  MHz.    By  the  end 
of  the  second  gate,  two  data  words  have  been 
loaded  into  the  counters.    These  two  4-digit 
words,  corresponding  to  Tj  and  T2  in  eq.  (3), 
provide  all  the  information  required  for  calcu- 
lating velocity  in  the  specimen  (relative  to 
water).    The  maximum  count  of  9999  corresponds 
to  a  round-trip  travel  time  through  500  mm  of 
water. 

A  10-step  hard-wired  program  sequentially  con- 
trols the  operation  of  a  4-function  calculator 
chip.    These  steps  include  clearing  the  calcu- 
later,  entering  the  values  Ti  and  T?,  performing 
the  arithmetic  operations  of  eq.  (3),  and  dis- 
playing the  result  to  4  significant  digits  via 
a  front-panel  LED  display.    In  the  event  of  a 
flag  generated  by  the  data  acquisition  circuitry, 
the  program  treats  T2/T1  as  a  negative  number. 
If  desired,  the  results  of  each  measurement  are 
printed  by  an  IBM  output  writer,  10  measurements 
to  a  line. 

The  timing> circuitry  maintains  synchronization 
among  all  elements  of  the  velocity  measuring  sys- 
tem.   When  the  user  presses  the  start  button 
after  checking  on  the  oscilloscope  that  the  cor- 
rect echoes  are  present,  the  timing  circuit  in- 
sures that  only  a  single  set  of  data  words  enters 
the  calculator  until  calculation  and  printout  of 
the  results  are  complete.    Using  a  free-running 
start  signal,  a  new  velocity  value  is  calculated 
and  printed  every  two  seconds. 


4.    Experimental  Results 
A.    Measurements  of  Velocity  in  Solids 

Before  conducting  measurements  of  ultrasound 
velocity  in  tissues,  it  was  necessary  to  check  the 
accuracy  of  the  reference  reflector  velocity  tech- 
nique and  its  implementation  in  the  device  de- 
scribed in  the  previous  section.    For  this  pur- 
pose three  objects  were  chosen:    a  4-mm  thick 
slab  of  polyvinylchloride  (PVC),  a  6-mm  thick 
slab  of  perspex,  and  a  10-mm  thick  combination  of 
the  two  materials.    The  ultrasonic  frequency  was 
5  MHz.    Fifty  velocity  measurements  were  made  for 
each  object. 

Results  of  measurements  of  velocity  in  the 
three  test  objects  are  given  in  table  1.  Absolute 
values  of  velocity  are  found  by  multiplying  the 
respective  relative  values  by  the  ultrasound 
velocity  in  water,  1498  m/s.    The  standard  devia- 
tions vary  from  about  0.3  to  0.5  percent  of  the 
measured  values. 


Table  1.    Velocity  measurements  of  selected  solids. 

Polyvinylchloride  (PVC) 

Relative  velocity  data  {-  10"') 


486 

1486 

1479 

1472 

1486 

1479 

1479 

1486 

1486 

1486 

486 

1486 

1500 

1500 

1486 

1486 

1479 

1479 

1486 

1479 

486 

1486 

1500 

1486 

1486 

1500 

1486 

1486 

1486 

1500 

486 

1500 

1486 

1486 

1486 

1486 

1472 

1479 

1500 

1479 

500 

1486 

1486 

1486 

1486 

1472 

1472 

1486 

1486 

1486 

Average  relative  velocity:  1.4860 
Average  absolute  velocity:    2226.0  m/s 
Standard  deviation;    11.2  m/s 

Perspex 


Relative  velocity  data 

10- 

1822 

1811 

1811 

1811 

1811 

1811 

1811 

1822 

1822 

1811 

1811 

1820 

1811 

1811 

1822 

1811 

1811 

1820 

1811 

1811 

1811 

1820 

1820 

1800 

1811 

1822 

1811 

1811 

1808 

1800 

1822 

1811 

1811 

1800 

1820 

1811 

1811 

1811 

1822 

1820 

1811 

1811 

1820 

1820 

1811 

1811 

1831 

1811 

1820 

1811 

Average  relative  velocity:  1.8138 

Average  absolute  velocity:    2717.1  m/s 
Standard  deviation:    9.4  m/s 

PVC-perspex  combination 

Relative  velocity  data  («  10" 3) 

1670    1664    1664    1668    1674    1668  1674    1668    1668  1664 

1664    1658    1664    1668    1664    1664  1670    1670    1674  1664 

1658    1664    1664    1664    1664    1664  1664    1674    1658  1664 

1658    1664    1674    1674    1670    1674  1668    1664    1664  1664 

1668    1668    1668    1664    1674    1658  1674    1658    1668  1668 

Average  relative  velocity:  1.6664 

Average  absolute  velocity:    2496.3  m/s 
Standard  deviation;    7.2  m/s 


To  check  the  accuracy  of  the  results  obtained 
for  each,  object,  the  thickness  given  by  eqs.  (4) 
and  (5)  was  compared  to  the  thickness  measured  by 
a  caliper.    Propagation  times  were  respectively 
3.60  ps,  4.42  us,  and  8.02  us  in  the  PVC,  perspex, 
and  PVC-perspex  combination.    The  corresponding 
thicknesses  calculated  from  the  data  and  eqs.  (4) 
and  (5)  are  4.003  mm,  6.005  mm,  and  10.01  mm, 
respectively,  which  are  within  0.1  percent  of  the 
measured  values. 


55 


B.    Measurements  of  Velocity  in  Soft  Tissues 

In  order  to  demonstrate  measurement  of  velocity 
in  soft  tissue,  a  sample  of  fat  in  vitro  and  two 
human  forearms  in  vivo  were  examined.    As  previ- 
ously, the  insonifying  frequency  was  5  MHz  and  50 
individual  velocity  measurements  were  taken  for 
each  target.    Results  of  the  measurements  are 
given  in  table  2. 

Table  2.    Velocity  measurements  of  soluble  soft  tissues. 
Fat  (bovine) 

Relative  velocity  data  (>  10"') 

0987  0987  0987  0987  0987  0987  0987  0987  0991  0991 

0991  0991  0991  0987  0987  0991  0987  0987  0987  0991 

0987  0987  0987  0991  0987  0991  0991  0991  0987  0987 

0991  0987  0991  0987  0991  0991  0987  0991  0983  0991 

0991  0987  0987  0987  0991  0987  0987  0987  0987  0987 

Average  relative  velocity:  0.9884 
Average  absolute  velocity;    1480.7  m/s 
Standard  deviation:    3.1  m/s 

Forearm  (subject  A) 

Relative  velocity  data  (»  10"^) 

1057    1058    1058    1058    1057    1057    1058    1057    1057  1056 

1056  1057    1057    1057    1057    1057    1058    1057    1058  1058 

1057  1057    1057    1057    1056    1057    1057    1057    1057  1057 

1056  1057    1056    1056    1057    1057    1057    1057    1057  1057 

1057  1058    1057    1059    1057    1057    1057    1057    1056  1056 

Average  relative  velocity:  1.0570 
Average  absolute  velocity:    1583.4  m/s 
Standard  deviation:    1.0  m/s 

Forearm  (subject  B) 

Relative  velocity  data  (■  10"') 

1056  1056  1055  1055  1056  1056  1055  1056  1055  1055 

1056  1055  1055  1056  1056  1055  1056  1056  1056  1055 

1055  1056  1055  1055  1056  1055  1054  1053  1054  1053 

1053  1053  1053  1054  1054  1055  1054  1054  1054  1053 

1054  1054  1054  1053  1054  1053  1053  1054  1053  1053 

Average  relative  velocity:  1.0546 
Average  absolute  velocity:    1579.8  m/s 
Standard  deviation:    1.7  m/s 


that  velocity  measurements  accurate  to  within 
0.2  percent  can  be  achieved  with  the  device. 

An  improved  model  is  now  under  construction. 
It  will  incorporate  microprocessor-based  circuit- 
ry for  increasing  the  number  of  measurements  at 
a  single  location  to  500  per  second.    This  will 
increase  the  precision  of  the  measured  velocity 
as  well  as  eliminate  problems  of  motion  during 
measurement.    The  new  device  should  enable  scan- 
ning a  structure  such  as  the  breast  with  the 
object  of  studying  local  variations  in  propaga- 
tion velocity. 

Acknowledgments 

The  author  wishes  to  express  his  gratitude  to 
the  Harry  de  Jur  Foundation  and  the  Rose  Teitel- 
baum  Cancer  Research  Foundation  for  their  par- 
tial support  of  this  research.    The  author  would 
also  like  to  thank  Professor  E.H.  Frei,  and  Y. 
Dreier,  I.  Gonen,  E.  Grinwald,  A.  Kuprak,  J. 
Leibovitz,  and  M.  Moshitzky  of  the  Weizmann 
Institute  for  their  roles  in  the  project. 

References 

[1]  Wells,  P.  N.  T.,  Absorption  and  dispersion 
of  ultrasound  in  biological  tissue.  Ultra- 
sound in  Med,  and  Biol.  U  369-376  (1975). 

[2]    Kossoff,  G.,  Fry,  E.  K. ,  and  Jellins,  J., 
Average  velocity  of  ultrasound  in  the  human 
female  breast,  J.  Acoust.  Soc.  Am.  53 , 
1730-1736  (1973T: 

[3]    Kossoff,  G.,  Reflection  Techniques  for  Mea- 
surement of  attenuation  and  velocity,  in 
Ultrasonic  Tissue  Characterization,  M.  Linzer, 
ed.,  National  Bureau  of  Standards  Spec.  Publ. 
453,  pp.  135-139  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 


The  specimen  of  bovine  fat  was  refrigerated 
for  several  days  and  then  warmed  to  room  tempera- 
ture before  insoni fi cation .    The  velocity  of 
1481  m/s  is  approximately  1  percent  higher  than 
the  velocity  for  cow  fat  given  by  Goldman  and 
Hueter  [4],  but  well  within  the  limits  for  mam- 
malian fat  in  general.    The  standard  deviation 
in  the  present  measurement  is  about  0.2  percent. 

The  next  measurements  were  made  in  the  fore- 
arms of  two  male  subjects,  the  same  age  (27). 
Care  was  taken  to  investigate  the  same  portion 
of  forearm  of  each  subject.    The  average  veloci- 
ties in  the  two  subjects,  1583  m/s  and  1580  m/s, 
respectively,  lie  within  the  range  given  by 
Goldman  and  Hueter  for  human  limbs.    The  standard 
deviations  are  of  the  order  of  0.1  percent  of  the 
measured  velocities.    More  data  would  be  required 
to  determine  if  the  difference  in  velocity  be- 
tween the  two  subjects  is  statistically  signifi- 
cant. 


[4]    Goldman,  D.  E.,  and  Hueter,  T.  F.,  Tabular 
data  on  the  velocity  and  absorption  of  high 
frequency  sound  in  mammalian  tissues,  J^. 
Acoust.  Soc.  Am.  28,  35-78  (1956). 


5.  Conclusion 


A  device  for  measuring  ultrasonic  propagation 
velocity  in  tissue  using  a  reference  reflector 
has  been  described.    On  the  basis  of  experimental 
results  in  solids  and  soft  tissue,  it  appears 


56 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


MEASUREMENT  OF  THE  TEMPERATURE  DEPENDENCE  OF  THE  VELOCITY 
OF  ULTRASOUND  IN  SOFT  TISSUES 


T.  Bowen,  W.  G.  Connor,  R.  L.  Nasoni  ,  A.  E.  Pifer,  and  R.  R.  Sholes 

Departments  of  Physics  and  Radiology 
University  of  Arizona 
Tucson,  Arizona    85724,  U.S.A. 


The  velocity  of  5  MHz  ultrasound  is  being  measured  in  tissue  samples  in  order 
to  evaluate  the  feasibility  of  non-invasive  monitoring  of  temperature  distributions 
produced  during  hyperthermia  treatments.    By  employing  a  pul sed-ul trasound  tech- 
nique in  which  the  time  of  the  first  zero-crossing  of  the  received  signal  is  re- 
corded, the  velocity  measurements  are  insensitive  to  reflections  and  to  changes  of 
attenuation.    Results  in  the  35  to  45  °C  range  for  fresh  canine  tissues,  such  as 
kidney,  liver,  and  muscle,  indicate  that  the  rate  of  velocity  change  with  tempera- 
ture is  correlated  with  the  magnitude  of  the  ultrasonic  velocity,  but  the  relation- 
ship appears  to  be  altogether  different  for  tissue  fat.    The  results  give  encourage- 
ment to  carry  out  the  more  extensive  measurements,  particularly  in  vivo,  which  would 
be  needed  to  determine  feasibility  of  an  ultrasonic  temperature  monitoring  system. 

Key  words:    Non-invasive  temperature  monitoring;  soft  tissue;  temperature  dependence; 
ultrasound  velocity;  zero  crossing  detection. 


1.  Introduction 

There  has  been  increasing  interest  recently 
in  hyperthermia  as  a  promising  modality  for  con- 
trol of  some  types  of  malignancies  [1-3]^.  In 
such  treatments  the  tissue  temperature  must  be 
raised  from  the  usual  body  temperature  of  37  °C 
to  approximately  43  °C  for  a  period  of  one-half 
to  one  hour.    If  the  temperature  is  accidentally 
a  few  degrees  higher,  all  cells  will  be  killed; 
if  a  few  degrees  lower,  the  effect  of  the  treat- 
ment will  be  negligible.    A  non-invasive  tem- 
perature monitoring  system  is  needed  for  this 
application.    In  addition,  such  a  system  might 
find  other  diagnostic  applications  where  thermal 
anomalies  are  caused  by  changes  in  the  circula- 
tory system  or  in  metabolic  processes. 

An  ultrasonic  temperature  monitoring  system 
might  be  devised  for  soft  tissue  if  some  ultra- 
sonic characteristic  varies  in  a  known  way  with 
temperature.    The  most  promising  parameters  ap- 
pear to  be  ultrasonic  velocity  and  attenuation. 
Earlier  work  on  transmission-type  tomographic 
image  reconstruction  systems  [4]  indicated  that 
velocity  data  gives  much  more  reproducible  re- 
sults.   However,  the  task  of  utilizing  velocity 
changes  to  measure  temperature  is  not  an  easy 
one.    At  37  °C  the  velocity  of  sound  in  water 
changes  1.8  m/s  per  °C  or  about  0.12  percent 
per  °C.    It  is  clear  when  one  examines  results 
for  the  velocity  of  ultrasound  in  any  particu- 


iFigures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


lar  tissue  type  that  the  variation  from  one 
sample  to  another  is  much  larger  than  the 
variation  over  a  10  °C  temperature  range;  con- 
sequently, one  cannot  hope  to  measure  tem- 
perature ultrasonically.    However,  it  still 
might  be  possible  to  measure  temperature 
'"hanges  ultrasonically,  making  use  of  the  fact 
it  the  initial  reference  temperature,  37  °C, 
is  known  to  be  quite  uniform  and  corrections 
for  regional  differences  are  also  known. 

Suppose  it  is  desired  to  monitor  a  tempera- 
ture change  from  body  temperature  (37  °C)  to 
a  treatment  temperature  in  the  neighborhood 
of  43  °C  within  ±  10  percent  for  each  tissue 
type  in  the  region  of  interest.    It  is  the 
purpose  of  this  work  to  discuss  the  first 
experimental  step  toward  answering  whether  or 
not  this  property  of  soft  tissue  can  be  pre- 
dicted with  such  accuracy. 

2.    Experimental  Arrangement 

The  experimental  arrangement  for  the  in  vitro 
measurement  of  ultrasonic  velocity  in  1  to  10  cm 
thick  tissue  samples  as  a  function  of  tempera- 
ture is  shown  schematically  in  figure  1.  The 
pulse  generator  provides  a  pulse  which  shock 
excites  the  12.7  mm  diameter  5  MHz  ultrasound 
transmitter  immersed  in  the  water  temperature 
bath.    This  pulse  also  starts  the  digital  timer. 
The  oscilloscope  serves  only  to  visually  moni- 
tor the  system;  its  trace  is  initiated  by  a 
trigger  pulse  from  the  pulse  generator  whose 
timing  can  be  delayed  or  advanced  relative  to 
the  main  pulse  so  as  to  observe  any  part  of 


57 


Digital  VOM 
(Keithley  168) 


Plastic 
Sample  Bag 


Thermistor 

 a 


Acoustic 
Transmitter 


r: 


Acoustic 
Receiver 


'  Heater/ 
Circulator 
(Haake  E52) 

Thermometer 


Tissue  Sample 


Temperature  Bath  Tank 


Pulse  Generator 
(HP  2I4A) 

Pulse  Trigger 
Out  Out 


Oscilloscope 
(Tektronix  R(vl45A) 

Ext  Vertical 
Trigger  Input 


(a)  Large  Signal 


Fig.  1.  Block  diagram  of  the  experimental 
arrangement  for  in  vitro  measure- 
ments of  ultrasound  velocity  as  a 
function  of  temperature. 

the  signal  at  the  receiver.    The  output  of  the 
12.7  mm  diameter  ultrasound  receiver,  which  is 
also  immersed  in  the  temperature  bath,  is  con- 
nected both  to  the  vertical  oscilloscope  input 
for  observation  of  its  amplitude  and  to  a  zero- 
crossing  discriminator  whose  output  stops  the 
digital  timer.    The  digital  timer  measures  to 
the  nearest  0.01  ms  the  transit  time  plus  a 
fixed  time  related  to  the  delays  in  the  system 
not  connected  with  the  sound  propagation.  The 
fixed  delays  are  found  by  an  extrapolation  to 
zero  separation  of  a  series  of  measurements 
of  transit  time  in  water  as  a  function  of 
separation  within  the  errors  of  measurement. 
This  agrees  with  the  results  of  others  who  have 
found  that,  although  the  pressure  distribution 
in  the  near  field  is  very  complicated,  the 
average  pressure  sensed  by  a  detector  whose 
dimensions  are  large  compared  to  a  wavelength  be- 
haves like  a  plane  wave  [5].    Adjustment  of  the 
width  of  the  shock  excitation  pulse  from  the 
pulse  generator  can  provide  a  signal  at  the  re- 
ceiver in  which  only  one  or  two  cycles  have  large 
ampl i  tude . 

The  zero-crossing  discriminator  is  an  essen- 
tial component  to  define  a  time  reference  in  a 
received  signal  burst.    Its  operation  is  illus- 
trated in  figure  2.    By  initiating  discriminator 
action  at  the  occurrence  of  the  first  zero 
crossing,  the  timing  becomes  independent  of  re- 
ceived signal  amplitude.    This  is  important  when 
studying  the  temperature  dependence  of  sound 
velocity  as  the  attenuation  (hence,  the  received 


receiver 
signal 


time 


first  zero  crossing 


(b)  Small  Signal 


recei ver 
signal 


time 


first  zero  crossing 


Fig.  2.  The  first  zero-crossing  is  shown 
for  a  large  signal  (a)  and  a  small 
signal  (b)  to  illustrate  that  its 
position  in  time  remains  unaffeced 
by  changes  of  amplitude. 


signal  amplitude)  also  varies  with  temperature. 
A  pulsed  ultrasound  system  with  the  timing  ref- 
erenced to  the  first  zero  crossing  not  only  is 
very  stable  in  operation  [6],  but  also  is  com- 
paratively immune  to  the  effects  caused  by 
multiple  reflections,  as  these  only  disturb  a 
later  part  of  the  received  signal. 

The  frequency  dependence  of  ultrasonic  veloc- 
ity and  attenuation  in  tissue  affects  the  shape 
of  the  received  signal,  which  could  introduce 
timing  shifts.    The  change  of  velocity  with  fre- 
quency, or  dispersion,  causes  the  pressure  pulse 
to  propagate  with  a  group  velocity  differing 
slightly  from  the  phase  velocity  of  a  continu- 
ous wave.    Tissue  dispersion  is  typically  0.1 
to  1.0  m/s  per  MHz  [7],  which  at  5  MHz  shifts 
the  group  velocity  0.5  to  5  m/s  above  the  phase 
velocity.    For  sample  thicknesses  on  the  order 
of  2  cm  the  first  zero  crossing  is  still  propa- 
gating with  a  velocity  nearly  equal  to  the 
group  velocity,  and  only  the  leading  edge  of  the 
first  one-half  cycle  is  distorted  because  the 
pulse  is  moving  faster  than  its  constituent  fre- 
quency components  [8].    A  small  velocity  shift 
of  this  magnitude  would  have  little  effect  upon 
the  observed  rate  of  change  with  temperature. 

The  increasing  attenuation  at  higher  fre- 
quencies in  tissue  changes  the  signal  waveform 
with  changing  depth.    However,  the  shape  of  the 
received  waveform  was  not  observed  to  change 
significantly  over  the  35  to  45  °C  range  of 
temperatures  at  fixed  depth. 

For  path  lengths  on  the  order  of  2  cm  the 
transit  time  jitter  is  less  than  0.001  \is .  The 
0.01  \iS  least  count  of  the  digital  timer  pres- 
ently limits  the  timing  precision,  but  this  can 
be  greatly  improved  with  the  same  time  digi- 
tization by  adding  an  event  counter  and  digital 
logic  so  that  the  transit  time  is  automatically 
accumulated  for,  say,  1000  transmitted  pulses. 
However,  this  improvement  will  come  about  only 
if  the  time  reference  oscillator  is  not  synchro- 
nized with  the  start  of  the  transit  time  meas- 
urement. 


58 


For  samples  more  than  a  few  centimeters 
thick,  which  may  often  be  encountered  for  i n 
vivo  measurements ,  the  optimum  ultrasound  fre- 
quency may  be  somewhat  less  than  the  5  MHz  ini- 
tially employed.    If  V(t)  is  the  received 
transducer  voltage  as  a  function  of  time  near 
the  first  zero  crossing  and  if  the  voltage  er- 
ror in  detecting  the  true  zero  point  is  given, 
it  is  desired  to  maximize  dV/dt  at  the  zero 
crossing.    This  would  be  proportional  in  the 
near  field  to  f  exp  (-gaf),  where  a  is  the 
spacing  between  transducers  and  gf  is  the 
ultrasonic  amplitude  attenuation  coefficient 
for  the  tissue  being  examined.    This  function 
is  maximum  when  gaf  =  1,  or  f  =  (ga)-  .  For 
example,  6  ^  0.01  ps-mm"!  is  typical  for  many 
soft  tissues,  so  if  a  =  100  mm,  f  =  1  MHz. 
This  result  can  be  easily  generalized  for  cases 
where  the  voltage  error  or  noise  is  frequency 
dependent. 

In  a  typical  data-taking  run,  the  tempera- 
ture of  the  bath  is  increased  at  a  rate  of  0.4 
°C/miri.    The  rate  must  be  reasonably  high  so 
that  changes  do  not  take  place  in  the  tissue 
which  are  a  function  only  of  time.  However, 
the  higher  the  rate  of  change  of  temperature, 
the  greater  the  lag  of  temperature  at  the 
center  of  the  sample.    In  order  to  correct  the 
data  for  this  temperature  lag,  a  thermistor  is 
inserted  at  the  center  of  the  sample  to  monitor 
the  temperature  difference  with  respect  to  the 
bath.    It  can  be  shown  from  the  solution  of  the 
one-dimensional  heat-diffusion  equation  when 
the  temperature  at  the  surfaces  is  a  linear 
function  of  time  that  the  average  temperature 
along  the  ultrasound  path  at  time  t,  Q(t),  is 
approximately  given  by: 

^(t)  =i«bath        ^  Wmistor  ' 

"^^'^  %ath  thermistor  ^^^^^^^  ^he 

temperatures  at  time  t  of  the  ,bath  (and  sample 
surfaces)  and  center  thermistor,  respectively. 


TEMPERATURE  (°C) 


Fig.  3.  The  speed  of  ultrasound  as  a  function  of 
temperature  for  water.  The  points  repre- 
sent data  obtained  in  this  work.  The 
solid  curve  represents  the  results  of  J. 
R.  Lovett,  J.  Acoust.  Soc.  Amer.  45, 
1051  (1969). 


For  samples  more  than  a  few  centimeters  thick 
and  for  in  vivo  measurement,  other  methods  of 
heating  the  sample  may  be  necessary  to  avoid 
large  temperature  inhomogeneities.    This  may 
often  necessitate  synchronized  alternate  puls- 
ing of  the  heat  source  and  the  ultrasound 
transmission  to  avoid  spurious  electrical 
pickup. 

Data  obtained  with  the  system  outlined  above 
for  water  is  illustrated  in  figure  3.    Note  the 
excellent  agreement  when  the  rate  of  variation 
with  temperature  is  compared  with  results  ob- 
tained by  others  [ 9] . 

3.  Results 

Freshly  excised  samples  of  canine  skeletal 
muscle,  liver,  kidney,  spleen,  brain,  and  fat 
were  measured  within  one  hour  after  sacrifice 
of  the  animal.    Each  run  began  at  approximate- 
ly 35  °C.    When  approximately  45  °C  was  reached 
the  temperature  was  held  constant  for  at  least 
30  minutes  to  verify  that  the  observed  changes 
were  attributable  to  the  changes  of  tempera- 
ture and  not  to  tissue  degeneration.    Some  data 
were  taken  where  the  temperature  was  held  con- 
stant at  other  values  between  35  and  45  °C;  no 
runs  gave  evidence  of  ultrasound  velocity 
changes  due  to  tissue  degeneration  for  samples 
within  the  first  hour  after  sacrifice.  At 
later  times  (>  60  minutes  after  sacrifice), 
velocity  changes  (occasionally  dramatic  de- 
creases) at  constant  temperature  were  ob- 
served with  muscle  samples.    Each  data  run  was 
least-square  fitted  to  a  polynomial  in  tem- 
perature containing  linear  and  quadratic  terms. 

Table  1.    Ultrasound  velocity  at  37  °C  and  rate 
of  change  of  ultrasound  velocity  with 
temperature,  dv/de,  at  37,  40,  43  °C 
for  fresh  canine  tissue  samples,  water, 
and  corn  oi 1 . 

Sound 

velocity       (dv/do) [ (m/s )/°C] 
Tissue        at  37  °C 


type 

(m/s 

37 

°C 

40  °C 

4; 

° 

Skeletal  muscle 

1589 

1 

1 . 

23 

1 

03 

0. 

82 

Skeletal  muscle 

1603 

3 

1 . 

13 

0 

92 

0. 

71 

Skeletal  muscle 

1588 

8 

1 . 

16 

0 

95 

0. 

78 

Skeletal  muscle 

1591 

6 

1 . 

08 

0 

87 

0 

65 

Liver 

1591 

7 

0. 

93 

0 

78 

0 

62 

Liver 

1594 

8 

1 

13 

0 

96 

0 

80 

Liver 

1604 

0 

0 

99 

0 

72 

0 

46 

Kidney 

1570 

2 

1 

35 

1 

16 

0 

98 

Kidney 

1566 

5 

1 

29 

1 

11 

0 

93 

Kidney 

1571 

1 

1 

29 

1 

11 

0 

94 

Bra  i  n 

1563 

2 

0 

67 

0 

62 

0 

26 

(white  matter) 

Spl een 

1601 

3 

1 

31 

1 

07 

0 

84 

Water 

1522 

5 

1 

84 

1 

66 

1 

48 

Stomach  fat 

1411 

9 

-2 

89 

-2 

85 

-2 

86 

(fresh) 

91 

Stomach  fat 

1412 

9 

-3 

43 

-2 

86 

-2 

.(refrig.  5  h) 

Corn  oil 

1420 

0 

-2 

75 

-2 

58 

59 


1600 


1550 


skeletal 
•muscl e 
'1  i  ver 


•  ••kidney 

•  brain 


•water 


1400 


1350 


stomach 
fat 


37       39        41  43 
TEMPERATURE  (°C) 


45 


Fig. 


Typical  data  showing  the  dependence 
of  ultrasound  velocity  on  tempera- 
ture for  fresh  canine  skeletal  muscle, 
liver,  kidney,  brain  (white  matter), 
and  stomach  fat,  as  well  as  for  water. 


The  results  are  listed  in  table  1  along  with 
the  results  for  water  and  corn  oil  for  compari- 
son.   Typical  curves  of  ultrasound  velocity 
versus  temperature  are  shown  in  figure  4.  In 
every  case,  the  variation  of  ultrasound  veloci- 
ty with  temperature,  dv/de',  is  less  than  the 
corresponding  value  for  water,  actually  re- 
versing sign  for  fat.    For  tissues  other  than 
fat  at  a  specified  temperature,  a  negative 
correlation  between  the  magnitude  of  the  ultra- 
sound velocity,  v,  and  dv/de  is  apparent.  This 
is  shown  in  figure  5,  where  dv/de  at  37  °C  is 
plotted  against  v  at  37  °C  for  each  sample  (ex- 
cept brain  and  fat)  and  for  water.    Except  for 
one  data  point  for  liver  and  the  single  point 
for  spleen  in  figure  5,  there  appears  to  be  a 
close  connection  between  dv/de  and  v  for  water 
and  soft  tissues  having  high  water  content. 


4.  Conclusions 

The  results  of  this  study  are  encouraging  in 
two  respects:    (1)    Changes  of  ultrasound  veloci- 
ty in  soft  tissues  with  temperature  can  be  simp- 
ly, yet  accurately,  measured  with  readily  avail- 
able digital  electronic  circuitry.    (2)    The  rate 
of  change  of  ultrasound  velocity  in  soft  tissues 
with  high  water  content  in  this  study  fell  with- 
in ±  10  percent  of  a  prediction  based  only  upon 


1.9 
1 .8 

1.7 

o 

3.  1.6 
1  1.5 

1.4- 
3  1.3 

1.2 

1.1 

1.0 

0.9 

1520 

Fig.  5. 


□  water 
A  kidney 
o  liver 
•  muscle 
X  spleen 


A 

A  A 


O 

J  I 


1540 


1560  1580 
v  (m/s) 


1600 


The  rate  of  change  of  ultrasound 
velocity  with  temperature,  dv/de, 
at  37  °C  is  plotted  as  a  function 
of  the  corresponding  ultrasound 
velocity,  v,  at  37  °C  for  each 
fresh  canine  tissue  sample  and  for 
water. 


the  magnitude  of  the  velocity  itself  for  9  out  of 
12  samples.    However,  it  is  already  apparent  that 
the  relative  fat  content  of  various  tissues  may 
be  an  important  parameter.    Before  any  conclu- 
sions can  be  reached  concerning  the  feasibility 
of  non-invasive  ultrasonic  temperature  monitor- 
ing, further  measurements,  especially  in  vivo, 
must  more  precisely  ascertain  the  predictability 
of  ultrasonic  velocity  temperature  dependence  in 
soft  tissues. 

References 

[1]  Gerner,  E.  W.,  Connor,  W.  G.,  Boone,  M.  L. 
M.,  Doss,  J.  D. ,  Mayer,  E.  G.,  and  Miller, 
R.  C,  Radiology  116,  433  (1975). 

[2]    Miller,  R.  C,  Connor,  W.  G.,  Heusinkveld, 
R.  S.,  and  Boone,  M.  L.  M.,  Prospects  for 
Hyperthermia  in  Human  Cancer  Therapy,  Part 
I:    Hyperthermic  Effects  in  Man  and  Spon- 
taneous Animal  Tumors  (to  be  published). 

[3]    Pettigrew,  R.  T.,  Gait,  J.  M.  ,  Ludgate, 
CM.,  Horn,  D.  B.,  and  Smith,  A.  N., 
Br.  J.  Surg.  61,  727  (1974). 

[4]    Greenleaf,  J.  F.  and  Johnson,  S.  A., 

Algebraic  Reconstruction  of  Spatial  Dis- 
tributions of  Acoustic  Velocity  and  At- 
tenuation in  Tissue  from  Time-of-Fl ight 
and  Amplitude  Profiles,  in  Ultrasonic  Tis- 
sue Characterization,  M.  Linzer,  ed.. 
National  Bureau  of  Standards  Spec.  Publ. 
453,  pp.  109-119  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 


60 


Rschevkin,  S.  N.,  The  Theory  of  Sound, 
pp.  440-444  (Macmillan,  New  York,  1963). 

Lees,  S.,  Gerhard,  F.  B.  Jr.,  and  Oppen- 
heim,  F.  G.,  Ultrasonics  11,  269  (1973). 

Carstensen,  E.  L.  and  Schwan,  H.  P.,  J_. 
Acoust.  Soc.  Amer.  3i,  305  (1959);  Carsten- 
sten,  E.  L.,  Absorption  of  Sound  in  Tissue 
(this  publication,  p.  29). 


[8]    Icsevgi ,  A.  and  Lamb,  W.  E.  Jr.,  Phys .  Rev. 
185,  517  (1969).    Section  IV  of  this  paper 
gives  an  elegant  and  clear  explanation  of 
the  behavior  of  the  pulse  velocity  when  the 
group  velocity  exceeds  the  phase  velocity. 

[9]    Lovett,  J.  R.,  J.  Acoust.  Soc.  Amer.  45, 
1051  (1969). 


61 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ULTRASONIC  ATTENUATION  IN  NORMAL  AND  ISCHEMIC  MYOCARDIUM 


M.  O'Donnell,  J.  W.  Mimbs,  B.  E.  Sobel ,  and  J.  G.  Miller 

Washington  University 
St.  Louis,  Missouri    63130,  U.S.A. 


The  ultrasonic  attenuation  coefficient  of  dog  myocardium  was  measured  in  vitro 
over  the  frequency  range  2  to  10  MHz.    Changes  in  the  attenuation  of  normal  myocardium 
were  measured  as  a  function  of  time  after  excision  at  fixed  temperatures.    Results  of 
measurements  made  on  tissue  maintained  at  35  °C  revealed  progressive  changes  in  the 
attenuation  as  a  function  of  time,  presumably  indicative  of  tissue  degradation.  Re- 
sults of  measurements  at  19.5  °C,  however,  showed  no  significant  changes  in  attenuation 
up  to  4  hours  following  excision.    The  temperature  dependence  of  the  attenuation  was 
measured  over  the  range     20  ^C  to     37  °C,  yielding  the  result  that  the  attenuation 
coefficient  at  37  °C  is  about  20  percent  lower  than  that  at  20  °C.    The  attenuation  co- 
efficient was  measured  in  vitro  at     20  °C  in  hearts  from  dogs  previously  subjected  to 
coronary  occlusion  and  sacrificed  at  intervals  ranging  from  15  minutes  to  3  days  fol- 
lowing occlusion.    Results  of  these  measurements  indicate  a  modest  decrease  in  the 
attenuation  of  ischemic  tissue  measured  at  15  minutes,  1  hour,  6  hours  and  24  hours, 
and  an  increase  in  attenuation  of  ischemic  regions  studied  3  days  following  occlusion. 

Key  words:    Ischemic  injury;  myocardial  infarction;  ultrasonic  attenuation. 


1.  Introduction 

Previous  reports  from  our  laboratory  presented 
the  ultrasonic  attenuation  in  normal  myocardium 
and  in  myocardium  subjected  to  ischemic  injury 
from  dogs  sacrificed  4  to  11  weeks  after  coronary 
artery  occlusion.    In  this  report,  we  present  the 
results  of  a  new  series  of  attenuation  measure- 
ments on  dog  left  ventricle.    Results  from  this 
series  of  measurements  are  analyzed  to:  (1) 
contrast  attenuation  in  normal  and  ischemic  zones 
of  tissue  from  animals  sacrificed  15  minutes, 
1  hour,  6  hours,  24  hours,  and  3  days  following 
coronary  occlusion,  (2)  examine  changes  in  the 
ultrasonic  attenuation  of  tissue  in  vitro  at  fixed 
temperatures  as  a  function  of  time  after  excision, 
and  (3)  examine  the  temperature  dependence  of  the 
ultrasonic  attenuation  of  tissue  in  vitro. 

2.  Methods 

A.    Ultrasonic  analysis 


TRANSMITTER 


SPECIMEN  RECEIVER 


DRIVER 


TIMING  UNIT 


SPECTRUM 
ANALYZER 


GATE 


SAMPLE 
a  HOLD 

ADC 

ADC 

DIGITAL 
PROCESSING 


STORAGE 


Fig.  1.    Block  diagram  of  the  ultrasonic 

instrumentation  used  in  this  study. 


The  instrumentation  employed  for  ultrasonic 
analysis  is  depicted  in  figure  1.    The  transmit- 
ting transducer  and  driver  yielded  an  ultrasonic 
pulse  with  frequency  components  of  sufficient 
amplitude  to  permit  operation  over  a  range  of 
2  to  10  MHz.    Details  of  transducer  design  are 
discussed  subsequently.    Under  control  of  a  tim- 
ing unit,  broadband  ultrasonic  pulses  were  gated 
into  a  slowly  sweeping  analogue  spectrum  analyzer. 
Output  pulses  from  the  spectrum  analyzer,  compris- 
ing the  logarithm  of  the  Fourier  transform  of  the 
received  ultrasonic  pulses,  were  converted  by  a 
sample-and-hold  unit  into  a  slowly  varying  (dc) 
signal.    As  the  spectrum  analyzer  was  slowly 


swept  through  frequency,  the  output  of  the  sample- 
and-hold  and  a  voltage  corresponding  to  the  fre- 
quency being  analyzed  by  the  spectrum  analyzer  were 
transmitted  to  two  analogue-to-digital  converters 
(ADC's).    The  digital  outputs  from  the  ADC's  were 
recorded  on  magnetic  tape  for  subsequent  analysis. 

The  procedure  used  for  the  quantitative  measure- 
ment of  the  attenuation  following  transmission  of 
ultrasound  through  tissue  has  been  described  in 
detail  in  previous  reports  [1,2]^  and  is  similar 
to  a  substitution  technique  developed  by  Schwan 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


53 


and  Carstensen  [3].    If  reflections  at  saline- 
tissue  interfaces  can  be  neglected  (and  in  the 
absence  of  artifacts  arising  from  phase  cancella- 
tion effects),  the  ultrasonic    attenuation  coeffi- 
cient      of  tissue  of  thickness  Az  is  given  by 

[Va(v)  -  Vb(v)]  ,  . 

at  =   ;   ^  ' 

^      AZ  log^oe 

where  the  term  [V/\(v)  -  Vb(v)]  represents  the 
signal  loss,  which  is  obtained  by  subtracting  the 
logarithmic  output  of  the  spectrum  analyzer  in 
the  presence  of  tissue  [Vd(v)]  from  that  recorded 
in  the  absence  of  tissue  IVa(v)].    The  signal  loss 
(expressed  in  decibels)  is  calibrated  by  comparison 
with  an  electronic  attenuation  standard. 

Figure  2  illustrates  the  use  of  the  method  to 
determine  the  attenuation  coefficient  of  a  test 
object,  a  flat  polyethylene  plate  of  Az  =  0.5  cm. 
The  left  panel  of  figure  2  exhibits  a  plot  of 
Va(v)  with  no  specimen  present  (upper  curve)  and 
a  plot  of  Vb(v)  with  the  polyethylene  specimen 
present  (lower  curve).    The  result  of  subtraction 
of  the  two  curves  in  the  left  panel  of  figure  2  is 
displayed  as  the  attenuation-frequency  plot  in  the 
right  panel,  with  data  points  calculated  according 
to  eq.  (1).    For  cases  in  which  reflections  from 
saline-specimen  interfaces  cannot  be  neglected, 
the  slope  of  the  attenuation  coefficient  versus 
frequency  curve  displayed  in  figure  2  remains  a 
valid  ultrasonic  index,  although  the  attenuation 
coefficient      may  not.    (The  ultrasonic  pulse 
length  used  in  these  studies  was  less  than  0.05  cm 
in  a  polyethylene  plate  of  Az  =  0.5  cm.  Thus 
standing  wave  effects  could  not  occur.) 

A  possible  source  of  artifact  in  the  apparent 
attenuation  coefficient       as  determined  from  eq. 
(1)  arises  from  phase  cancellation  effects  [1,2]. 
Phase  cancellation  effects  may  occur  if  inhomo- 
geneities  in  the  tissue  distort  the  ultrasonic 
field  presented  to  a  spatially  extended  piezo- 
electric receiving  transducer.    These  wavefront 
distortions  may  result  from  transmission  of  ultra- 
sound through  tissue  with  variations  in  surface 
characteristics,  internal  structural  characteris- 
tics, or  both.    When  the  wavefronts  incident  upon 
a  piezoelectric  receiver  are  distorted,  the 


generated  electrical  signal  is  degraded  because 
of  the  phase  sensitive  nature  of  the  receiver. 
Phase  cancellation  effects  can  induce  artifacts 
into  attenuation  data  which  might  be  interpreted 
incorrectly  as  reflecting  only  the  absorption  and 
scattering  properties  of  a  specimen  [1,2,4]. 

An  intensity  sensitive  ultrasonic  receiving 
transducer  which  is  inherently  insensitive  to 
phase  cancellation  effects  is  under  continuing 
development  in  our  laboratory  [1,2].    This  trans- 
ducer makes  use  of  the  acoustoelectric  effect  in 
single  crystal  cadmium  sulfide  to  convert  the  in- 
tensity of  an  incident  ultrasonic  wave  into  an 
electronic  current  in  the  semiconducting  crystal. 
An  illustration  of  phase  cancellation  effects  is 
presented  in  figure  3,  which  contrasts  the  per- 
formance of  a  piezoelectric  and  an  acoustoelectric 
transducer  of  equal  diameters  (1.3  cm).    The  upper 
panel  of  figure  3  illustrates  the  apparent  attenu- 
ation coefficient  as  a  function  of  frequency  for 
four  adjacent  sites  of  dog  left  ventricle  mea- 
sured with  the  piezoelectric  receiver.  Marked 
variability  is  observed  in  data  obtained  from  four 
morphologically  similar  regions.    The  lower  panel 
depicts  the  results  of  measurements  on  the  identi- 
cal four  sites;  however,  the  intensity  sensitive 
acoustoelectric  receiver  was  used.    The  data  of 
the  lower  panel  exhibit  more  uniform  consistency, 
presumably  reflecting  the  reduction  or  elimination 
of  artifacts  due  to  phase  cancellation. 

The  present  configuration  of  the  acousto- 
electric receiver  is  cumbersome  and  would  not 
permit  the  rapid  collection  of  data  that  was  re- 
quired in  experiments  on  ischemic  tissue  (see 
below).    All  data  reported  here  were  obtained 
using  identical  focused  piezoelectric  transducers 
(1.3  cm  diameter,  5  cm  focal  distance)  for  trans- 
mitter and  receiver.    By  positioning  the  specimen 
of  interest  so  that  it  lay  entirely  within  the 
depth  of  field  of  the  focused  transducers,  phase 
cancellation  artifacts  were  minimized.    A  com- 
parison of  the  performance  of  the  focused  trans- 
ducers and  the  acoustoelectric  receiver  is  shown 
in  figure  4.    In  this  figure  the  apparent  at- 
tenuation coefficient  versus  frequency  plots  ob- 
tained from  analysis  of  the  identical  four  ad- 
jacent sites  of  left  ventricle  are  presented. 
The  upper  panel  presents  data  obtained  with  the 


Fig.  2.    The  left  panel  is  a  plot  of  Va(v)  obtained  without  a  specimen  and  Vb(v)  obtained 
with  a  specimen  (a  polyethylene  plate  of  Az  =  0.5  cm  in  this  case).    The  right 
panel  depicts  the  attenuation  versus  frequency  plot  determined  using  eq.  (1)  and 
the  data  shown  in  the  left  panel. 


64 


0.8 

0.6 

0.4 

0.2- 

0.0 
1.0 

0.8  - 

0.6  - 

0.6 

0.4 


Piezoelectric  receiver 


Acoustoel ectric  receiver 


O.OL 


4  6 
FREQUENCY  (MHz) 


Fig.  3.    Planar  piezoelectric  transmitter.  The 
upper  panel  depicts  the  apparent  attenu- 
ation coefficient  as  a  function  of  fre- 
quency for  four  contiguous  sites  of  dog 
left  ventricle  measured  with  a  1.3  cm 
diameter  planar  piezoelectric  receiver. 
The  lower  panel  depicts  the  results  of 
measurements  on  the  identical  four  sites 
using  a  1.3  cm  diameter  intensity  sensi- 
tive acoustoel ectric  receiver. 


0.8r 


0.6 


0.4 


0.2 


0.0 


0.8 


0.6 


0.4 


0.2 


Focused  piezoelectric 
receiver 


0.0 


Acoustoelectric  receiver 


4  6 
FREQUENCY  (MHz) 


Fig.  4.    Focused  piezoelectric  transmitter.  The 
upper  panel  illustrates  the  apparent  at- 
tenuation coefficient  as  a  function  of 
frequency  for  four  contiguous  sites  of  dog 
left  ventricle  measured  using  a  1.3  cm  di- 
ameter focused  (5  cm  focal  distance)  piez- 
oelectric receiver.    The  lower  panel  de- 
picts the  results  of  measurements  on  the 
identical  four  sites,  using  a  1.3  cm  diam- 
eter acoustoelectric  receiver. 


focused  piezoelectric  receiver  and  the  lower 
panel  presents  data  obtained  with  the  acousto- 
electric receiver.    A  focused  piezoelectric  trans- 
mitter was  used  to  obtain  the  results  shown  in 
both  panels  of  figure  4.    The  focused  piezo- 
electric transducer  arrangement  (fig.  4)  is  seen 
to  offer  significant  improvement  over  the  planar 
piezoelectric  transducer  arrangement  of  figure  3. 
Data  reproducibility  and  consistency  remain, 
however,  more  satisfactory  with  the  use  of  the 
acoustoelectric  receiver. 


In  an  attempt  to  quantitate  the  extent  to 
which  phase-cancellation  effects  resulted  in  de- 
gradation of  attenuation-frequency  data,  a 
statistical  index  was  developed.    Because  phase 
cancellation  effects  in  attenuation-frequency 
data  may  manifest  themselves  as  erratic  (non- 
monotonic) frequency  dependences  (in  contrast  with 
the  expected  monotonic  variation  of  attenuation 
with  frequency),  one  method  of  segregating  the 
data  is  on  the  basis  of  a  statistical  test  of  the 
goodness  of  a  fit  to  a  monotonic  curve.    Using  a 


65 


three  parameter  fit  to  the  measured  signal  loss 
date  [(V/\-Vb)  of  eq.  (1)],  a  root  mean  square 
deviation  (RMSD)  was  calculated.    The  RMSD  was 
defined  by 


RMSD  =  (v,  -  "0''  jV   [ym(-^')  -  yc(v)]  j  dv  (2) 

where  v  is  the  ultrasonic  frequency,        -  v-^)  de- 
fining the  frequency  interval  of  the  measurements. 
y^{v)  is  the  measured  signal  loss  at  frequency  v, 
and  y(-(v)  is  the  calculated  signal,  loss,  deter- 
mined by  a  least  squares  three  parameter  fit  to 
the  data,     ihe  units  of  RMSD  are  the  same  as  those 
of  y(v),  in  this  case  decibels  (dB).    The  value  of 
RMSD  was  used  to  quantitate  the  degree  of  non- 
monotonicity  and  thus  to  estimate  in  a  relative 
fashion  the  contribution  of  phase  cancellation  ef- 
fects.   Large  values  of  RMSD  would  be  expected 
whenever  substantial  phase  cancellation  contribu- 
tions influenced  the  apparent  attenuation. 

The  application  of  this  approach  is  demonstrat- 
ed in  figure  5,  which  illustrates  our  progress  in 
systematically  reducing  artifacts  due  to  phase 
cancellation.    The  calculated  RMSD  for  each  at- 
tenuation-frequency curve  obtained  with  several 
transducer  combinations  is  shown.    Data  obtained 
in  20  measurements  with  a  1.3  cm  diameter  planar 
piezoelectric  receiver  are  shown  in  panel  A  and 
in  44  measurements  using  a  0.2  cm  diameter  planar 
piezoelectric  receiver  in  panel  B.    In  both  cases 
the  transmitter  was  a  1.3  cm  diameter  planar 
piezoelectric  transducer.    The  reduced  RMSD's  of 


1  3cm  diameter  A 
1                                 piezoelectric  receiver 
N=  20 

/ 

\                                 0  2  cm  diameter  n 

piezoelectric  receiver 
1                                 N  --  44 

1 

1                               Focused  piezoeleclric  r- 
,                                 receiver  and  tionsmitter 

1                                    N  =  42 

1                                 1.3cm  diameter 

'                                 acoustoelertf ic  receiver  D 

1                                       N  :  33 

1             ill             1             1             1             1             1  1 

00        05         10         15         20        2.5        30        35        4  0  45 


ROOT  MEAN  SQUARE  DEVIATION  (dB) 

Fig.  5.    Histogram  showing  the  range  of  cal- 
culated root  mean  square  deviations 
(RMSD's)  of  attenuation-frequency 
curves  determined  with  four 
different  transducer  configurations. 

panel  B  as  compared  with  panel  A  presumably  re- 
flect a  decrease  in  the  phase  cancellation  arti- 
facts achieved  by  reducing  the  area  over  which 
the  ultrasonic  field  was  integrated.    Panel  C 
exhibits  the  results  of  measurements  on  42  sites 
made  with  a  focused  transducer  pair  and  demon- 
strates substantial  improvement  over  the  arrange- 


ments of  panels  A  and  B.     (It  should  be  noted, 
however,  that  substantial  artifacts  due  to  phase 
cancellation  occur  in  the  presence  of  intervening 
tissue  located  outside  of  the  focal  zone.  Thus, 
for  future  applications  in  vivo,  the  use  of  a 
focused  transducer  does  not  appear  to  be  a  satis- 
factory method  for  minimizing  artifacts  due  to 
phase  cancellation.)    The  results  of  the  measure- 
ments on  33  sites  using  the  acoustoelectric  re- 
ceiver are  presented  in  panel  D  of  figure  5.  The 
very  small  RMSD's  obtained  with  the  intensity 
sensitive  acoustoelectric  receiver  reflect  the 
inherent  freedom  from  phase  cancellation  arti- 
facts, making  a  transducer  of  this  design  poten- 
tially suitable  for  future  in  vivo  applications. 

B.    Tissue  preparation  and  independent 
indices  of  pathology 

All  measurements  were  carried  out  in  vitro  on 
myocardial  tissue  obtained  from  adult,  mongrel, 
15  to  30  kg  dogs.    In  studies  of  ischemic  injury, 
each  dog  was  intubated  after  anesthesia  with 
sodium  pentobarbital  (25  mg/kg,  intravenously), 
placed  on  a  Harvard  respirator,  ventilated  with 
room  air,  and  subjected  to  a  left  thoracotomy  via 
the  fifth  interspace.    The  pericardium  was  in- 
cised, and  the  left  anterior  descending  coronary 
artery  dissected  free  immediately  distal  to  the 
first  ventricular  branch  and  ligated.    The  peri- 
cardium was  left  open.    For  dogs  studied  at  6 
hours,  24  hours,  and  3  days  following  occlusion, 
the  chest  was  closed  conventionally,  with  intra- 
pleural suction  maintained  for  at  least  one  hour 
via  a  chest  tube. 

At  the  time  of  sacrifice,  each  animal  was  again 
anesthetized  with  sodium  pentobarbital  (50  mg/kg, 
intravenously).    The  heart  was  rapidly  excised 
and  placed  in  a  0.9  percent  NaCl  solution.  To 
prepare  the  tissue  for  ultrasonic  analysis  an  in- 
cision was  made  at  the  root  of  the  aorta,  con- 
tinued interiorly  along  the  left  ventricular  sur- 
face of  the  interventricular  septum  to  the  apex, 
and  extended  posteriorly  and  superiorly,  terminat- 
ing at  the  base  of  the  heart.    This  permitted 
prompt  excision  of  a  segment  of  the  left  ventric- 
ular anterior  and  apical  wall.    This  segment  com- 
prised the  area  of  infarction  and  surrounding 
area  of  normal  myocardium  in  all  animals  subject- 
ed to  coronary  ligation.    (We  note  that  this  seg- 
ment was  measured  unaltered,  i.e.,  no  additional 
preparation  such  as  further  slicing  to  achieve 
flat  or  parallel  surfaces  was  undertaken.)  Except 
as  noted,  ultrasonic  analysis  was  initiated  within 
5  minutes  following  the  death  of  the  animal,  and 
completed  within  45  minutes. 

Tissue  from  animals  sacrificed  at  24  hours  and 
72  hours  following  coronary  occlusion  was  analyzed 
biochemically  for  creatine  kinase  (CK)  content, 
an  established  index  of  tissue  pathology.  One 
centimeter  diameter  biopsies  corresponding  to  re- 
gions of  ultrasonic  analysis  were  obtained  im- 
mediately following  ultrasonic  measurement.  To 
facilitate  comparison  of  data  from  different  ani- 
mals, regional  CK  activity  was  expressed  as  the 
percentage  of  depletion  compared  to  activity  in 
normal  myocardium  from  the  same  animal.  Sites 
with  CK  depletion  of  greater  than  40  percent  were 
classified  as  ischemic  and  those  with  CK  depletion 
less  than  20  percent  were  classified  as  normal. 

For  dogs  sacrificed  from  15  minutes  to  6  hours 
following  occlusion,  colloidal  carbon  black  was 


66 


injected  into  the  left  atrium  fifteen  seconds 
prior  to  killing  the  animal  for  the  purpose  of 
differentiating  regions  of  ischemic  and  normal 
myocardium.    Regions  of  ischemia  did  not  change 
color  after  injection,  while  non-ischemic  regions 
rapidly  were  stained  black.    Visual  inspection  of 
the  tissue  after  excision  was  used  to  differen- 
tiate normal  from  ischemic  tissue.    Analysis  of 
60  sites  from  4  control  dogs  indicated  that  the 
presence  of  dye  did  not  alter  the  ultrasonic 
measurements  of  normal  myocardium. 

3.  Results 

Prior  to  addressing  questions  related  to 
ischemic  injury,  it  is  necessary  to  determine  ex- 
perimentally the  range  of  variability  introduced 
into  the  results  of  ultrasonic  attenuation  mea- 
surements in  vitro  by  variables  such  as:    (1)  the 
time  interval  that  elapses  between  the  sacrifice 
of  the  dog  and  the  ultrasonic  measurement,  (2) 
the  temperature  of  the  tissue  during  ultrasonic 
measurement,  and  (3)  the  specific  region  of  the 
left  ventricle  that  is  analyzed. 

A.    Time  interval  between  sacrifice 
and  measurement 

In  an  effort  to  estimate  the  extent  to  which 
tissue  degradation  compromises  the  results  of  in 
vitro  measurements,  the  ultrasonic  attenuation 
coefficient  of  dog  myocardium  was  measured  at 


Time 
after 
exci  sii 


15  min 
2  h 
4  h 


15  min 
2  h 
4  h 


B.    Temperature  dependence  of  the  attenuation 

To  relate  the  results  obtained  at  20  °C  to 
what  might  be  expected  if  the  measurements  were 
carried  out  at  37  °C,  an  additional  series  of 
experiments  was  designed  to  elucidate  the  tem- 
perature dependence  of  the  attenuation  in  a 
manner  relatively  independent  of  effects  aris- 
ing from  tissue  degradation.    Results  from  the 
previous  series  of  experiments  (table  1)  sug- 
gested the  possibility  that  tissue  might  be 
temporarily  stored  at  approximately  20  °C 
without  undergoing  significant  degradation  be- 
tween measurements  made  at  elevated  temperatures. 
Using  this  approach,  four  freshly  excised  hearts 
were  used  for  a  measurement  of  the  attenuation  at 
20.5  °C,  25  °C,  30  °C,  and  37  °C.    Each  segment 
of  left  ventricle  was  placed  in  a  20.5  °C  bath 
and  analyzed  ul trasoni cally  immediately  after 


constant  temperature  as  a  function  of  time  fol- 
lowing excision.    All  experiments  were  performed 
in  a  0.9  percent  NaCl  bath  with  temperature  re- 
gulated to  ±  0.2  °C  by  means  of  a  controller  and 
reci  rcul ator. 

In  table  1  data  are  displayed  as  a  function  of 
time  following  excision  for  experiments  performed 
at  two  temperatures:    (1)  19.5  °C  (analysis  of  27 
regions  from  3  hearts)  and  (2)  35  °C  (analysis  of 
27  regions  of  3  additional  hearts).    The  ultra- 
sonic attenuation  coefficient  as  a  fraction  of 
frequency  was  approximately  independent  of  time 
up  to  4  hours  following  excision  for  tissue  main- 
tained at  19.5  °C.    When  the  temperature  was 
maintained  at  35  °C,  however,  a  definite  change 
with  time  was  observed.    For  measurements  at 
35  °C,  the  value  of  the  slope  of  a  least  squares 
line  fit  to  the  attenuation  versus  frequency  data 
changed  from  (0.061  ±  0.003)  cm-iMHz-i  (mean  ± 
standard  error  (S.E.))  at  15  minutes  following 
sacrifice  to  (0.071  +  0.003)  cm-iMHz-i  at  4 
hours  following  sacrifice.    It  thus  appears  that 
no  statistically  significant  changes  in  attenua- 
tion occur  for  periods  up  to  several  hours  fol- 
lowing excision  for  tissue  maintained  at  approxi- 
mately 20  °C,  although  small  {■'^  15  percent)  but 
significant  changes  occur  over  that  time  interval 
for  tissue  maintained  at  35  °C.    These  results 
motivated  the  choice  of  20  °C  as  the  temperature 
at  which  the  extensive  series  of  measurements  on 
ischemic  and  normal  myocardial  tissue  in  vitro 
reported  below  were  carried  out. 


excision.    After  analysis  at  this  temperature, 
the  tissue  sample  and  holder  were  removed  from 
the  measurement  bath  and  placed  in  a  storage  bath 
which  was  maintained  at  19.5  °C.    The  measurement 
bath  was  heated  to  25  °C  and  regulated  to  maintain 
this  temperature  to  ±  0.2  °C.    Upon  stabilization 
of  the  temperature  at  25  °C,  the  sample  and  hold- 
er were  reinserted  into  the  original  bath  and  al- 
lowed to  equilibrate,  whereupon  the  ultrasonic 
attenuation  was  measured.    The  same  procedure  was 
repeated  for  measurements  at  30  °C  and  37  °C, 
with  the  entire  process  lasting  about  80  minutes. 
To  check  the  validity  of  this  method  for  identify- 
ing the  true  temperature  dependence,  the  attenua- 
tion obtained  at  37  °C  using  the  above  procedure 
was  compared  to  the  attenuation  measured  at  37  °C 
for  3  additional  freshly  excised  hearts  at  27 
sites.    The  slope  of  the  attenuation  for  these 
measurements  made  at  37  °C  within  minutes  of  ex- 


Table  1.    Ultrasonic  attenuation  of  normal  myocardium  as  a  function  of  time  after  excision. 


Number  Number  Temp.                 Attenuation  coefficient  (cm"')  (mean  +  SE)  Slope  of  a  vs^ 

of       of   ■  frequency 


sites 

dogs 

O 

C 

2 

MHz 

4 

MHz 

6 

MHz 

8 

MHz 

10 

MHz 

( 

cm  ' 

MHz 

-1) 

27 

3 

19 

.5 

0. 

10 

±  0 

02 

0. 

19 

±  0 

02 

0 

34 

±  0 

02 

0 

38 

±  0.03 

0 

65 

±  0 

03 

0 

072 

±  0 

002 

27 

3 

19 

.5 

0 

10 

±  0 

02 

0. 

20 

±  0 

02 

0 

36 

±  0 

02 

0 

50 

±  0.03 

0 

70 

z  0 

03 

0 

075 

±  0 

002 

27 

3 

19 

.5 

0 

10 

±  0 

02 

0 

20 

±  0 

02 

0 

35 

±  0 

02 

0 

49 

±  0.03 

0 

69 

±  0 

03 

0 

.075 

±  0 

002 

27 

3 

35 

0 

09 

+  0 

02 

0 

17 

±  0 

03 

0 

30 

±  0 

03 

0 

43 

+  0.03 

0 

56 

±  0 

04 

0 

.061 

±  0 

003 

27 

3 

35 

0 

09 

±  0 

02 

0 

19 

±  0 

03 

0 

34 

±  0 

03 

0 

47 

±  0.03 

0 

61 

±  0 

04 

0 

.068 

±  0 

003 

27 

3 

35 

0 

11 

±  0 

02 

0 

22 

±  0 

.03 

0 

37 

±  0 

03 

0 

52 

±  0.03 

0 

64 

±  0 

04 

0 

.071 

±  0 

003 

67 


Table  2.    Ultrasonic  attenuation  of  normal  myocardium  as  a  function  of  temperature. 
Number  Number  Temp.    Attenuation  coefficient  (cm"^)  (mean  ±  SE)  Slope  of  a  vs 


of 
ites 

of 

dogs 

2 

MHz 

4 

MHz 

6 

MHz 

8 

MHz 

10 

MHz 

frequency 
(cm-'  MHz-i) 

36 

4 

20.5 

0 

10 

±  0 

02 

0 

19 

±  0 

02 

0 

34 

±  0 

03 

0 

48 

+  0 

03 

0 

64 

±  0 

03 

0.071  +  0.002 

36 

4 

lb 

U 

10 

±  0 

02 

0 

19 

+  0 

02 

0 

33 

±  0 

03 

0 

45 

±  0 

03 

0 

61 

±  0 

03 

0.058  ±  0.UU2 

36 

4 

30 

0 

10 

±  0 

02 

0 

17 

±  0 

02 

0 

31 

±  0 

03 

0 

43 

±  0 

03 

0 

58 

±  0 

03 

0.054  ±  0.002 

36 

4 

37 

0 

11 

±  0 

02 

0 

19 

±  0 

02 

0 

31 

±  0 

03 

0 

41 

±  0 

03 

0 

56 

±  0 

03 

0.058  ±  0.002 

cision  was  (0.060  ±  0.002)  cm"i  MNz"!  in  good 
agreement  with  the  value  0.058  ±  0.002  cm-^  MMz"! 
measured  at  37  °C  for  the  36  sites  of  tissue 
stored  at  19.5  °C  between  measurements.    This  re- 
sult suggests  that  the  procedure  is  methodologi- 
cally sound. 

Results  of  this  series  of  experiments  on  36 
regions  from  4  hearts  designed  to  determine  the 
temperature  dependence  of  the  attenuation  coeffi- 
cient are  presented  in  table  2.    The  slope  of  a 
least  squares  line  fit  to  the  attenuation- 
frequency  data  decreases  approximately  linearly 
with  increasing  temperature  over  the  range 
20.5  °C  to  37  °C,  with  the  value  at  37  °C  being 
about  20  percent  less  than  that  at  20.5  °C. 

C.    Regional  variations  in  attenuation 

In  order  to  identify  possible  variations  in  the 
attenuation  coefficient  as  a  function  of  the  re- 
gion of  the  left  ventricle  that  was  analyzed,  102 
sites  from  4  dogs  not  subjected  to  coronary  liga- 
tion were  studied  and  the  results  segregated  into 
4  groups  based  on  the  region  investigated.  Data 
from  these  experiments  are  presented  in  table  3. 
Values  of  the  attenuation  coefficient  for  regions 
in  the  mid-posterior  and  apical  aspects  of  the 
left  ventricle  were  essentially  identical.  Some- 
what higher  values  were  exhibited  by  regions  in 
the  papillary  muscles  and  generally  lower  values 
were  exhibited  by  regions  near  the  base  of  the 
heart.    These  results  are  summarized  in  figure  6, 
utilizing  the  slope  of  a  least  squares  line  fit 
to  the  attenuation  coefficient  versus  frequency 
plots  for  comparative  purposes. 

Based  on  the  results  of  this  study  of  regional 
variability  of  the  ultrasonic  attenuation  coeffi- 
cient of  normal  myocardium,  subsequent  studies 
were  conducted  primarily  on  sites  from  mid- 
posterior  and  apex,  avoiding  sites  from  the 
papillary  muscles  and  the  base.    With  this  ap- 


proach, the  value  of  the  slope  of  the  attenuation 
for  normal  left  ventricle,  as  determined  from 
measurements  at  245  discrete  sites  from  36  dogs, 
was  (0.072  ±  0.001)  cm"!  MHz'i  (mean  ±  S.E.). 


0.08 
iS  0.06 
0.04 

LU 

o  0.02 


posterior 
aspect 


=21 


papi 1 lary 
muscl es 


7^ 

// 

'// 

/' 

/  / 

f. 

N=38 

N=21 

N=21 

apex 

mid- 

base 

posterior 


Fig.  6.    Regional  variation  of  the  slope  of  the 
attenuation  (2  to  10  MHz)  in  normal 
myocardium. 


Number  Number  Temp 
Descrip-      of  of 
tion         sites     dogs  °C 


Table  3.    Regional  variation  of  ultrasonic  attenuation  in  normal  myocardium. 

Attenuation  coefficient  (cm"M  (mean  ±  SE) 


2  MHz 


4  MHz 


5  MHz 


8  MHz 


10  MHz 


Slope  of  g  vs 

frequency 
(cm'i  MHz-i) 


20  0.11  ±  0.02  0.20  ±  0.02  0.34  ±  0.02  0.45  ±  0.03  0.62  +  0.03  0.064  ±  0.002 
20    0.10  ±  0.02    0.19  ±  0.02    0.34  ±  0.02    0.48  ±  0.03    0.64  ±  0.03    0.070  ±  0.002 


Base 

Mid- 
posterior 

Apex 


21 
21 

39 


Papillary  21 
muscles 


20  0.10  ±  0.02  0.19  ±  0.02  0.35  ±  0.02  0.49  ±  0.03  0.65  ±  0.03  0.071  ±  0.002 
20     0.13  +  0.02    0.26  ±  0.02    0.41  ±  0.02    0.55  ±  0.03    0.76  ±  0.03    0.079  ±  0.002 


68 


D.    Effects  of  ischemic  injury 

To  investigate  changes  in  physical  properties  of 
tissue  resulting  from  ischemic  injury,  ultrasonic 
measurements  were  carried  out  on  freshly  excised 
hearts  previously  subjected  to  coronary  occlusion 
for  intervals  of  15  minutes,  1  hour,  6  hours,  24 
hours,  and  3  days.    Results  of  these  measurements 
are  summarized  in  table  4.    The  attenuation  meas- 
ured at  15  minutes,  1  hour,  6  hours,  and  24  hours 
following  coronary  occlusion  is  lower  than  that  ex- 
hibited by  normal  myocardium,  while  that  at  3  days 
following  occlusion  is  higher  than  that  of  normal. 

The  attenuation  values  reported  in  table  4  rep- 
resent averages  of  the  values  obtained  for  all 
sites  analyzed.    Results  illustrated  in  figures  4 
and  5  suggest  that  artifacts  arising  from  phase 
cancellation  effects  may  occur  in  data  obtained 
using  the  focused  transmitter-focused  receiver  con- 
figuration.   To  eliminate  measurements  exhibiting 
probable  phase  cancellation  effects  from  further 
consideration,  an  index  based  on  RMDS  (eq.  2)  was 
used  to  segregate  the  data.    Attenuation  coeffi- 
cient versus  frequency  curves  exhibing  RMSD's 
greater  than  0.25  dB  were  deleted  from  further  con- 
sideration.   The  value  0.25  dB  was  selected  for 


segregating  the  data  because  it  represents  an  RMSD 
two  times  the  maximum  RMSD  exhibited  by  data  ob- 
tained using  the  phase  cancellation  insensitive 
acoustoel ectric  receiver.    The  results  of  segre- 
gating the  data  in  this  way  are  illustrated  in 
table  5.    The  average  values  of  the  slope  of  the 
attenuation  versus  frequency  curves  are  presented 
for  two  groupTl     {^)  data  including  sites  exhibit- 
ing phase  cancellation  effects,  and  (2)  data  ex- 
cluding sites  exhibiting  phase  cancellation  effects. 
A  comparison  of  the  number  of  sites  contributing 
to  the  average  slope  in  each  group  indicates  that  a 
modest  but  significant  number  of  attenuation  versus 
frequency  curves  were  compromised  by  phase  cancel- 
lation effects. 

The  results  of  experiments  on  normal  and  ische- 
mic myocardium  are  summarized  in  figure  7.  The 
average  slope  of  the  attenuation  (last  column  of 
table  5)  is  shown  at  15  minutes,  1  hour,  6  hours, 
24  hours,  and  3  days  following  coronary  occlusion. 
The  slope  of  the  attenuation  of  ischemic  myocardium 
is  significantly  lower  than  that  of  normal  myocard- 
ium for  times  ranging  from  15  minutes  through  24 
hours  following  coronary  occlusion  and  is  signifi- 
cantly larger  at  3  days  following  occlusion. 


Descrip-  of 
tion  sites 


Table  4.    Ultrasonic  attenuation  of  normal  and  ischemic  myocardium 
Number  Number  Temp.  Attenuation  coefficient  (cm'^)  (rtiean  ±  SE) 

C 


of 

dogs 


2  MHz 


4  MHz 


6  MHz 


8  MHz 


10  MHz 


Slope  of  g  vs 

frequency 
(cm-'  MHz-') 


Normal 

Time  after 
occl usion 


245 


36 


20 


0.10  ±  0.02    0.19  ±  0.02    0.33  ±  0.02    0.48  ±  0.03    0.65  ±  0.03    0.072  ±  0.001 


15 

min 

56 

5 

20 

0 

10 

0 

02 

0 

17 

+ 

0 

02 

0 

32 

+ 

0 

02 

0 

43 

+ 

0 

03 

0 

57 

+ 

0 

03 

0 

063 

+ 

0 

003 

1 

h 

64 

8 

20 

0 

09 

+ 

0 

02 

0 

17 

0 

02 

0 

32 

4- 

0 

02 

0 

43 

+ 

0 

03 

0 

57 

+ 

0 

03 

0 

063 

+ 

0 

003 

6 

h 

59 

5 

20 

0 

08 

0 

02 

0 

15 

+ 

0 

02 

0 

30 

+ 

0 

02 

0 

44 

0 

03 

0 

57 

+ 

0 

03 

0 

066 

+ 

0 

002 

24 

h 

29 

11 

20 

0 

08 

0 

02 

0 

16 

+ 

0 

02 

0 

29 

+ 

0 

02 

0 

43 

0 

03 

0 

59 

+ 

0 

03 

0 

067 

+ 

0 

002 

3 

d 

31 

7 

20 

0 

11 

+ 

0 

02 

0 

20 

+ 

0 

02 

0 

38 

+ 

0 

02 

0 

52 

0 

03 

0 

69 

+ 

0 

03 

0 

079 

+ 

0 

002 

Table  5.    Slope  of  the  ultrasonic  attenuation  of  normal  and  ischemic  myocardium. 

Including  sites  Excluding  sites 

exhibiting  phase  exhibiting  phase 

Description  cancellation  effects        cancellation  effects 


Time  after         Numoer     Temperature    Character       Number  Slope  Number  Slope 

coronary  occlusion    of  dogs  °C  of  sites         of  sites    (cm"'  MHz"M     of  sites    (cm"^  MHz'M 


15 

min 

5 

20 

normal 

58 

0 

073 

± 

0 

003 

45 

0 

072 

+ 

0 

003 

15 

min 

5 

20 

ischemic 

56 

0 

063 

+ 

0 

003 

49 

0 

062 

+ 

0 

002 

1 

h 

8 

20 

normal 

69 

0 

072 

+ 

0 

003 

52 

0 

071 

0 

003 

1 

h 

8 

20 

i  schemic 

64 

0 

063 

+ 

0 

003 

57 

0 

062 

+ 

0 

003 

6 

h 

5 

20 

normal 

59 

0 

072 

+ 

0 

002 

51 

0 

072 

+ 

0 

002 

6 

h 

5 

20 

ischemic 

59 

0 

066 

+ 

0 

002 

49 

0 

064 

+ 

0 

002 

24 

h 

11 

20 

normal 

29 

0 

073 

+ 

0 

002 

24 

0 

068 

+ 

0 

002 

24 

h 

11 

20 

ischemic 

29 

0 

067 

+ 

0 

002 

27 

0 

060 

+ 

0 

002 

3 

d 

7 

20 

normal 

30 

0 

072 

+ 

0 

002 

25 

0 

072 

+ 

0 

001 

3 

d 

7 

20 

i  schemic 

31 

0 

079 

+ 

0 

002 

28 

0 

082 

0 

002 

£9 


I    0  08 

M 
X 


.0  07 


So  06 

o 


005 


Fig.  7. 


0.09 

i 

NORMAL 

ISCHEMIC 

N= 
24  27 


I5min  Ih  6h  24  h 

TIME  AFTER  OCCLUSION 


25  28 


3d 


Slope  of  the  ultrasonic  attenuation 
(2  to  10  MHz)  for  normal  and  ischemic 
myocardial  tissue  determined  at  specified 
intervals  following  coronary  occlusion. 


tenuation  might  be  expected  to  accompany  edema, 
which  is  a  recognized  correlate  of  recent 
ischemic  injury.    If  the  small  decrease  in  at- 
tenuation of  ischemic  tissue  is  corroborated  by 
measurements  ir[  vivo,  the  use  of  an  index  based 
on  the  ultrasonic  attenuation  to  characterize  the 
state  of  myocardium  might  be  approached.  Efforts 
to  use  an  index  based  on  attenuation  to  charac- 
terize the  state  of  myocardial  tissue  might 
utilize  each  heart  as  its  own  control  thus  abro- 
gating data  variability  due  to  regional  myocardial 
variations  as  documented  in  table  3. 

An  increase  in  attenuation  over  that  of  normal 
myocardium  is  observed  in  tissue  subjected  to 
ischemic  injury  and  measured  three  days  following 
coronary  occlusion.    Previous  reports  from  this 
laboratory  documented  substantial  increases  in  at- 
tenuation in  tissue  studied  4  to  11  weeks  after 
coronary  occlusion  [2,9].    This  increase  in  at- 
tenuation, evident  at  three  days  following  occlu- 
sion and  reaching  substantial  magnitude  weeks 
after  occlusion,  appears  to  be  related  to  the  on- 
set of  scar  formation,  reflecting  the  increase  in 
the  collagen  content  of  necrotic  tissue. 


4.  Discussion 

Emphasis  in  the  present  studies  was  placed  on 
systematic  measurements  making  use  of  carefully 
defined  control  experiments.    Either  creatine 
kinase  depletion  or  reduced  uptake  of  colloidal 
carbon  was  used  as  an  independent  index  for 
documenting  myocardial  ischemia.    Care  was  taken 
to  eliminate  artifacts  due  to  ultrasonic  phase 
cancellation  effects,  a  potentially  significant 
source  of  variability  in  attenuation  coefficient 
measurements.    Methods  suitable  for  conducting 
meaningful  experiments  on  excised  tissue  were  de- 
veloped and  validated.    Results  of  the  study 
addressing  the  problem  of  tissue  degradation  fol- 
lowing excision  led  to  the  choice  of  ^  20  °C 
rather  than  37  °C  for  the  present  series  of  in- 
vestigations.   The  possibility  of  significant 
changes  in  the  ultrasonic  properties  of  tissue 
measured  in  vitro  as  a  function  of  time  follow- 
ing excision  was  considered  by  Hueter  [5],  and 
more  recently  by  Frizell  and  Carstensen  [6],  and 
by  Bamber  et  al .  [7].    Previous  studies  examined 
changes  for  time  intervals  substantially  larger 
(tens  of  hours)  than  the  present  experiments, 
which  had  the  more  limited  goal  of  defining  the 
range  of  validity  of  attenuation  measurements  on 
freshly  excised  tissue. 

The  temperature  dependence  of  the  attenuation 
of  normal  myocardium  was  measured  over  the  range 
^.  20  °C  to     37  °C  using  a  technique  designed  to 
minimize  artifacts  arising  from  tissue  degrada- 
tion.   Results  of  these  measurements  suggest  that 
the  data  obtained  at  37  °C  might  be  expected  to 
differ  by  only  about  20  percent  from  that  ob- 
tained at  20  °C.    However,  these  limited  measure- 
ments on  normal  myocardium  do  not  adequately  ad- 
dress the  possibility  of  differences  in  the  tem- 
perature dependence  of  the  attenuation  exhibited 
by  ischemic  as  opposed  to  normal  tissue. 

Results  of  this  study  indicate  a  small  de- 
crease in  the  attenuation  in  the  early  stages  of 
a  myocardial  infarct  (up  to  24  hours  following 
coronary  occlusion).    Although  mechanisms  re- 
sponsible for  the  attenuation  of  soft  tissue  are 
inadequately  understood  [8],  a  decrease  in  at- 


Acknowl edgments 

Lawrence  J.  Busse  carried  out  the  investiga- 
tions of  phase  cancellation  effects  described  in 
the  text  and  presented  in  figures  3,  4,  and  5. 
Pranoat  Suntharothok-Priesmeyer  was  responsible 
for  production  of  the  text  and  illustrations. 

This  work  was  supported  in  part  by  grants 
HL19537,  HL17646,  and  HL07081  from  the  National 
Institutes  of  Health. 


References 

[1]    Busse,  L.  J.,  Miller,  J.  G.,  Yuhas,  D.  E., 
Mimbs,  J.  W.,  Weiss,  A.  N.,  and  Sobel ,  B.  E., 
Phase  Cancellation  Effects:    A  Source  of 
Attenuation  Artifacts  Eliminated  by  a  CdS 
Acoustoelectric  Receiver,  in  Ul trasound 
in  Medicine,  D.  White,  ed.,  vol.  3,  pp. 
1519-1535  (Plenum  Press,  New  York,  1977). 

[2]    Miller,  J.  G.,  Yuhas,  D.  E.,  Mimbs,  J.  W. , 
Dierker,  S.  B.,  Busse,  L.  J.,  Laterra,  J.  J., 
Weiss,  A.  N.,  and  Sobel,  B.  E.,  Ultrasonic 
Tissue  Characterization:    Correlation  Between 
Biochemical  and  Ultrasonic  Indices  of  Myo- 
cardial Injury,  in  Proceedings  1976  IEEE 
Ultrasonics  Symposium  79,  33-43  (Cat.  No. 
CM  1120-43U,  IEEE,  New  York,  1976). 

[3]    Schwan,  H.  P.  and  Carstensen,  E.  L.,  Ultra- 
sonics aids  in  diathermy  experiments. 
Electronics,  216-220  (July  1952). 

[4]    Marcus,  P.  N.  and  Carstensen,  E.  L.,  Problem 
with  absorption  methods  of  inhomogeneous 
solids,  J.  Acoust.  Soc.  Am.  58_,  1334-1335 
(1975). 

[5]    Hueter,  T.A.,  WADC  Tech.  Rept.  57-706  (1958). 

[6]    Frizell,  L.  A.,  Ultrasonic  Heating  of  Tissue, 
Ph.D.  Thesis,  University  of  Rochester  (1976) 
(unpublished). 


70 


[7]    Bamber,  J.  C,  Fry,  M.  J.,  Hill,  C.  R. ,  and 
,  Dunn,  F.,  Ultrasonic  attenuation  and  back- 
scattering  by  mammalian  organs  as  a  function 
of  time  after  excision,  Ultrasound  in  Medi- 
cine and  Biology  3,  15-20  TT977T- 

[8]    See  the  references  listed  in  O'Donnell,  M. 
and  Miller,  J.  G. ,  Mechanisms  of  Ultrasonic 
Attenuation  in  Soft  Tissue  (this  publication, 
p.  37). 

[9]    Yuhas,  D.  E.  ,  Mimbs,  J.  W.,  Miller,  J.  G., 
Weiss,  A.  N.,  and  Sobel  ,  B.  E.,  Changes  in 
Ultrasonic  Attenuation  Indicative  of  Regional 
Myocardial  Infarction,  in  Ultrasound  in 
Medicine,  D.  W.  White,  ed.,  vol.  3,  1883- 
1894  (Plenum  Press,  New  York,  1977). 


71 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ACOUSTIC  MICROSCOPIC  ANALYSIS  OF  MYOCARDIUM 


Donald  E.  Yuhas  and  Lawrence  W.  Kessler 

Sonoscan  Inc. 
720  Foster  Avenue 
Bensenville,  niinois    60106,  U.S.A. 


Acoustic  microscopy  can  be  employed  to  measure  variations  in  the  ultrasonic  attenua- 
tion and  velocity  over  spatial  dimension  of  tens  of  micrometers.    Knowledge  of  the  in- 
teractions at  this  fine-structure  level  provides  a  more  complete  understanding  of  the 
response  of  tissues  at  diagnostic  frequencies.    In  this  article  we  investigate  the  ultra- 
sonic characteristics  of  formalin-fixed  infarcted  and  normal  myocardium  at  the  micro- 
scopic level.    The  elastic  microstructure  of  normal  tissues  is  found  to  be  uniform  over 
spatial  dimensions  of  hundreds  of  micrometers.    The  attenuation  coefficient  a,  at 
100  MHz,  ranges  from  37  to  68  cm"^  while  the  velocity  varies  less  than  ±  20  m/s. 
Acoustic  anisotropy,  attributed  to  muscle  fiber  orientation,  is  found  to  be  a  primary 
source  of  attenuation  variability.    The  elastic  microstructure  in  the  zones  of  infarc- 
tion is  very  distinct  from  that  in  normal  zones.    With  infarction,  variations  in  the 
attenuation  ranging  from  38  to  100  cm"i  occur  over  dimensions  of  hundreds  of  micro- 
meters.   Accompanying  this  variability  in  attenuation  is  a  wide  distribution  of  velocity. 
The  most  highly  attenuating  regions  in  the  infarcted  tissue  show  localized  increases  in 
the  velocity  of  more  than  75  m/s.    Such  large  variations  in  sonic  velocity  are  sufficient 
in  magnitude  to  be  an  important  source  of  phase  cancellation  in  low  frequency  attenua- 
tion measurements.    Additionally,  the  100  MHz  data  indicate  that  the  frequency  depend- 
ence of  the  attenuation  coefficient  observed  in  the  1  to  10  MHz  range  for  infarcted  and 
normal  myocardium  may  not  hold  in  the  10  to  100  MHz  frequency  interval. 

Key  words:    Acoustic  microscopy;  anisotropy;  attenuation;  elastic  microstructure; 

infarct;  interferogram;  phase  cancellation;  myocardium;  velocity. 


1.  Introduction 

The  complete  ultrasonic  characterization  of 
tissue  can  be  divided  into  two  distinct  levels 
of  interaction:    1)  structures  greater  than  the 
wavelength  of  diagnostic  ultrasound  and  2)  struc- 
tures that  are  finer.    The  larger  structures 
give  rise  to  specular  reflections  which  are  used 
to  define  the  outlines  of  tissues  and  organs 
seen  in  conventional  B-scan  echograms.  The 
finer  structures  give  rise  to  characteristic 
scattering  properties  of  tissue  and  are  also 
responsible  for  intrinsic  attenuation  and  velo- 
city.   It  is  this  sensitivity  to  fine  structure, 
that  makes  the  diagnostic  frequency  ultrasonic 
parameters  so  important  in  determining  tissue 
pathologies. 

As  an  aid  to  understanding  the  low  frequency 
scattering,  attenuation,  and  velocity  data, 
direct  visualization  of  the  finer  structures  is 
desirable.    Conventional  microscopy  (optical  and 
electron)  can  be  used  to  delineate  various  struc- 
tural components,  but  these  methods  do  not  provide 
the  required  information  regarding  the  ultrasonic 
properties  of  the  tissue. 

On  the  other  hand,  the  elastic  microstructural 
information  obtained  through  acoustic  microscopy 
is  directly  applicable  to  the  lower  frequency  (1 
to  10  MHz)  characteristics.    Acoustic  microscopy 
provides  a  means  for  visualizing  directly  the 


elastic  architecture  as  well  as  for  measuring 
variations  in  the  ultrasonic  attenuation  and 
velocity  properties  over  areas  tens  of  micro- 
meters in  diameter.    By  using  this  technique, 
valuable  insight  can  be  gained  into  mechanisms 
responsible  for  the  attenuation  of  sound  as  well 
as  factors  which  influence  tissue  characteriza- 
tion measurements  at  lower  frequencies. 

This  preliminary  study  consists  of  a  series  of 
acoustic  microscope  observations  of  myocardial 
tissue.    Previous  reports  have  shown  that  quanti- 
tative changes  in  ultrasonic  attenuation  over  the 
frequency  range  2  to  10  MHz  are  associated  with 
regional  myocardial  infarction  [1-4]^.    The  ultra- 
sonic attenuation  in  normal  tissue  was  found  to 
exhibit  a  linear  frequency  dependence,  while  in- 
farcted tissue  exhibits  a  significant  quadratic 
dependence.    As  a  clue  to  elucidating  the  nature 
of  the  change  in  attenuation  at  lower  frequencies, 
acoustic  microscopy  was  used  to  investigate  the 
elastic  microstructure  of  normal  and  infarcted 
myocardi  urn. 

2.  Methods 

Acoustic  micrographs  were  obtained  with  a  com- 
mercially available  100  MHz  scanning  laser  acoustic 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


73 


microscope,  SONOMICROSrOPE  100.    A  brief  descrip- 
tion of  the  technique  is  presented  here  for  com- 
pleteness.   For  a  more  detailed  description,  the 
reader  may  refer  to  the  listed  references  [5-9]. 

Figure  1  illustrates  the  instrumentation  used 
to  produce  100  MHz  acoustic  micrographs.    A  sample 
is  placed  on  a  glass  stage  between  an  insonifying 
transducer  and  receiver.    The  transmitting  trans- 
ducer is  a  piezoelectric  element  which  is  bonded 
to  a  glass  substrate,  and  the  receiver  is  a  scan- 
ning focused  laser  beam.    The  acoustic  energy  is 
coupled  to  the  specimen  through  the  glass  sub- 
strate without  need  for  an  acoustically  lossy 
water  bath.    The  specimen  is  moistened  with  liquid 
to  ensure  good  acoustic  coupling.    The  sound  is 
transmitted  through  the  sample  at  an  angle  of  10° 
with  respect  to  the  normal.    In  order  to  provide 
a  specularly  reflecting  surface  for  the  scanning 
laser  beam  receiver,  the  specimen  is  covered  with 
a  partially  silvered  coverslip.    The  coverslip  al- 
lows a  small  amount  of  light  to  penetrate  to  the 
sample  and  render  an  optical  view  simultaneously 
with  the  acoustic. 


laser  beam  scanners 


imaging  optics 


demodulator  and 
photodetector 


mirrored  cover'sl  in  i  g_ 


spec \men 


acoustic 
signal  processor 


acoustic 
frequency  generator 


T 


stage 
photodetector 


ultrasonic  transducer 


optical 
signal  processor 


Acoustic  Image 
Display 


Fig.  1.    Schematic  diagram  of  the  acoustic  micro- 
scope used  in  this  study.    The  instrument 
employs  a  scanning  laser  beam  technique 
for  detecting  the  amplitude  and  phase  of 
acoustic  energy  transmitted  through  the 
sample.    Acoustic  and  optical  images  are 
produced  simultaneously  and  are  presented 
on  separate  monitors. 

The  acoustic  energy  transmitted  through  the 
specimen  imparts  a  slight  oscillatory  mechanical 
perturbation  to  the  coverslip  surface.    The  ampli- 
tude of  these  perturbations,  which  vary  at  the 
acoustic  frequency,  is  inversely  proportional  to 
the  localized  ultrasonic  attenuation  properties  in 
the  underlying  sample.    By  sensing  the  perturba- 
tions and  displaying  the  signal  on  a  TV  monitor, 
an  acoustic  image  is  formed.    Images  are  obtained 
in  real  time  and  displayed  at  a  magnification  of 
about  70  X.    The  bright  regions  seen  on  the 
acoustic  micrographs  correspond  to  areas  of  high 
transmission  through  the  sample,  whereas  the 
darker  areas  correspond  to  regions  of  higher  ultra- 
sonic attenuation. 

Quantitative  attenuation  measurements  of  the 
tissue  observed  in  the  microscope  can  be  made  any- 
where within  the  field  of  view  by  an  optical  tech- 


nique described  previously  [10].    With  the  sample 
in  place,  an  acoustic  image  is  obtained.    Areas  of 
interest  are  noted  and  readings  of  the  image  bright- 
ness are  obtained  with  a  light  meter.    Once  the 
image  brightness  is  recorded  the  sample  is  removed 
and  replaced  by  an  equivalent  thickness  of  coupling 
medium.    For  this  work  formalin  (10  percent  formal- 
dehyde) was  used.    Electrical  attenuation  is  then 
inserted  into  the  electro-acoustic  signal  path  to 
restore  the  image  brightness  to  its  previous  level. 
Although    the  acoustic  field  is  quite  uniform, 
care  was  taken  to  ensure  that  the  same  region  of 
sound  field  was  measured  for  the  cases  with  and 
without  the  sample  in  place.    Using  this  technique, 
attenuation  can  be  measured  over  an  area  as  small 
as  5  X  10"^  cm2.    The  sensitivity  of  the  photo  cell 
used  in  these  studies  permitted  discrimination  of 
sound  levels  to  within  ±  1  dB,  which  introduced  a 
3  to  8  percent  uncertainty  in  the  measured  at- 
tenuation coefficient.    Because  the  acoustic  image 
of  the  specimen  is  observed  during  the  course  of 
the  measurement  procedure,  it  is  possible  to  select 
regions  of  low  contrast  structural  details,  there- 
by assuring  the  attenuation  values  obtained  were 
free  from  artifacts  arising  from  vessels  or  fluid 
filled  cavities. 

In  addition  to  recording  the  transmissivity  of 
specimens,  the  scanning  laser  acoustic  microscope 
is  equipped  with  an  acoustic  interferogram  mode  of 
operation.    Acoustic  interferograms  display  the 
localized  distribution  of  transit  times  through 
the  sample.    For  specimens  of  uniform  thickness 
these  transit  time  variations  are  directly  related 
to  localized  variations  in  sonic  velocity.  Inter- 
ferograms can  be  conceptualized  with  reference  to 
figure  2.    In  this  schematic,  an  ultrasonic  plane 
wave  is  incident  on  the  sample  at  an  angle  with 
respect  to  the  normal.    The  intersection  of  the 
wavefronts  with  the  plane  below  the  specimen  re- 
sults in  a  series  of  fringes  (indicated  by  the 
dots)  which  run  in  and  out  of  the  plane  of  the 
page.    Using  phase  sensitive  detection,  the  posi- 
tions of  these  equiphase  lines  are  displayed  on 


elastic  plane  of 

inclusion  /  observation 


Fig.  2.    Model  for  conceptualizing  acoustic  inter- 
ferograms.   As  a  plane  wave  propagates 
through  a  sample,  regions  of  velocity 
variations  cause  refraction.    The  result- 
ing wavefronts  are  displaced  as  they  ar- 
rive at  the  sample  surface  as  shown  by 
the  dots.    The  lateral  position  of  the 
wavefronts  (fringes)  are  a  function  of  the 
change  in  velocity  and  can  be  calculated 
from  the  formulas  given  in  the  text. 


74 


the  image  monitor.    Figure  3  is  a  micrograph  of  an 
interferogram  recorded  with  no  specimen  on  the 
stage.    In  this  case  the  interferogram  lines 
(fringes)  are  straight  and  parallel.    Going  from 
left  to  right,  each  successive  interference  fringe 
represents  an  increase  in  transit  time  through  the 
glass  stage  of  exactly  one  period  of  oscillation 
(10  nanoseconds  at  100  MHz). 


Fig.  3.    Interferogram  obtained  eith  no  sample  on 
the  microscope  stage  showing  equally 
spaced,  parallel,  interference  fringes. 
This  is  the  same  type  of  interferogram 
that  would  be  obtained  on  a  sample  that 
exhibited  no  regional  changes  in  sonic 
veloci  ty. 

Figure  2  also  illustrates  the  effect  of  elastic 
inhomogeneity  on  the  position  of  the  interferogram 
fringes.    As  the  wave  enters  the  sample  it  refracts 
according  to  Snell's  law.    In  the  areas  of  the 
sample  with  uniform  propagation  velocity  (white 
areas)  the  separation  of  successive  fringes  after 
propagating  through  the  specimen  is  the  same  as 
that  measured  with  no  sample.    When  structures 
with  variations  in  propagation  velocity  are  present, 
illustrated  by  the  hatched  area,  the  separation  of 
successive  fringes  in  the  vicinity  of  the  elastic 
inclusion  changes.    The  magnitude  of  the  fringe 
shift,  N,  is  directly  related  to  the  propagation 
velocities  of  the  two  components  by  the  following 
relations  [11]: 

M  =  (cot  ei  -  cot  62)  (1) 

where  N  is  the  dimensionless  magnitude  of  the 
lateral  fringe  shift  normalized  by  the  unperturbed 
fringe  spacing:    Xi  is  the  wavelength  of  sound  in 
region  1  of  the  specimen;  At  is  the  specimen 
thickness;  and  ei  and  62  are  the  propagation  angles 
in  regions  1  and  2  respectively. 

The  relationship  between  the  propagation  angles 
and  the  sonic  velocity  is  given  by  Snell's  law: 

sin  9i  _  sin  62  ^2) 

Ci  C2 

where  Cj  and  C2  are  the  respective  propagation 
velocities. 

Using  eqs.  (1)  and  (2)  and  knowing  the  magni- 
tude of  the  sonic  velocity  of  one  component  in  the 
field  of  view,  the  sonic  velocities  of  all  other 


areas  can  be  determined.  The  primary  factor  de- 
termining the  sensitivity  of  the  technique  for 
measuring  variations  in  velocity  is  the  sample 
thickness.  For  the  sample  thicknesses  used  in 
this  study,  a  displacement  of  a  fringe  by  1/10  of 
the  unperturbed  fringe  spacing  (i.e.,  N  =  0.1) 
corresponds  to  a  velocity  change  of  5  m/s. 

3.    Myocardial  Samples 

The  specimens  used  in  this  study^  were  taken 
from  regions  of  formalin-fixed  canine  myocardium. 
Sections  .05  ±  .005  cm  in  thickness  were  made  in 
several  different  myocardial  regions  extending 
from  the  apex  to  the  base  of  the  heart.    The  plane 
of  dissection  was  oriented  approximately  at  an 
angle  of  45°  with  respect  to  the  long  axis  of  the 
heart  and  each  section  extended  from  epicardial  to 
endocardial  surface.    Samples  were  taken  from 
regions  of  infarcted  and  non-infarcted  tissues 
with  the  regions  of  infarction  lying  nearer  to  the 
apex  of  the  heart.    Analysis  was  carried  out  on 
section  taken  from  a  single  heart  which  had  been 
formalin  fixed  for  more  than  a  year  and  was 
originally  excised  4  weeks  after  coronary  occlusion. 

4.  Results 

Figure  4  presents  an  acoustic  micrograph  (a) 
and  interferogram  (b)  obtained  on  a  myocardial 
section  taken  from  a  non-infarcted  region.  Bright- 
ness variations  in  this  micrograph  (fig.  4a)  gen- 
erally range  from  light  to  dark  grey.    A  band  of 
lighter  material  approximately  1  mm  wide  can  be 
seen  running  almost  vertically  through  the  central 
portion  of  the  micrograph.    On  either  side  of  this 
acoustically  transmissive  region  are  darker  areas 
oriented  both  perpendicular  and  parallel  to  the 
band.    Except  for  the  dark  structure  (indicated  by 
the  arrow),  which  is  attributed  to  a  vessel,  the 
change  in  brightness  between  the  various  regions 
is  relatively  subdued  and  gradual.    The  interfero- 
gram of  this  region  (fig.  4b)  shows  a  uniform 
sonic  velocity  distribution.    Only  the  area  in- 
dicated by  the  arrow  exhibits  an  abrupt  disruption 
of  the  interferogram  lines  showing  a  slight  in- 
crease of  sonic  velocity  of  20  m/s.  Throughout 
most  of  the  sample  the  lateral  shifts  in  the  inter- 
ferogram lines  going  from  the  top  to  the  bottom  of 
the  micrograph  over  the  entire  field  are  less  than 
.4  fringes  indicating  velocity  changes  of  less 
than  ±  10  m/s  in  the  vertical  direction. 

Figure  5  illustrates  an  interesting  effect, 
that  of  acoustic  anisotropy  in  myocardial  tissue. 
These  micrographs  were  obtained  on  the  same  sample 
shown  in  figure  4;  however,  the  sample  has  been 
rotated  90°  in  the  plane  of  the  stage.  Recall 
that  the  angle  of  insonif ication  is  approximately 
10°  from  the  specimen  normal.    Thus,  while  the 
major  component  of  the  propagation  vector  K  is 
perpendicular  to  the  sample,  a  small  off-axis  com- 
ponent leads  to  some  dramatic  changes  in  image 
contrast  on  rotation.    The  amplitude  micrograph 
shown  in  figure  5a  reveals  two  distinct  regions. 
The  upper  half  of  the  micrograph  shows  acoustic 
morphology  similar  to  that  observed  in  the  pre- 
vious micrographs.    The  lower  half  of  the  micro- 


^Samples  provided  courtesy  of  Washington 
University,  Division  of  Cardiology,  St. 
Louis,  Missouri. 


75 


(a) 


Fig.  4.    Acoustic  amplitude  micrographs  (a)  and 
interferogram  (b)  of  normal  myocardium. 
The  field  of  view  in  each  micrograph  is 
2.3  by  3.0  mm  and  the  sample  thickness  is 
500  micrometers.    In  the  amplitude  micro- 
graph the  light  areas  are  regions  of  low 
attenuation.    The  interferograms  are  ar- 
ranged so  that  a  lateral  shift  to  the  right 
corresponds  to  a  region  of  increased  velo- 
city and  a  shift  to  the  left,  lower  velo- 
city.   A  shift  of  one  fringe  corresponds 
to  a  velocity  change  of  50  m/s.  Typical 
variations  in  this  sample  are  +  10  m/s. 


Fig.  5.    Acoustic  amplitude  micrograph  (a)  and 

interferogram  (b)  of  the  same  region  de- 
picted in  figure  4  after  sample  rotation 
(counter-clockwise)  by  90°.    The  dramatic 
changes  in  these  images  compared  to  figure 
4  are  due  to  the  tissue  being  anisotropic 
acoustical ly. 

The  top  portion  of  the  interferogram  shows 
velocity  variations  of  ±  5  m/s,  while 
variations  of  40  m/s  are  seen  in  the  lower 
portion  of  the  micrograph. 


graph  is  significantly  different,  showing  bands  of 
acoustically  light  and  dark  material  running  hori- 
zontally across  the  field  of  view.    The  change  of 
acoustic  contrast  is  abrupt  across  these  bounda- 
ries and  their  morphology  suggests  that  they  arise 
from  muscle  fiber  orientation.    In  this  micrograph 
the  off-axis  component  of  the  propagation  vector 
k  is  parallel  to  the  acoustic  striations  seen  in 
the  micrograph,  i.e.  with  the  grain,  whereas  in 
the  previous  figure  it  is  perpendicular  or  across 
the  grain. 

It  should  be  emphasized  here  that  the  portion 
of  the  sample  comprising  the  lower  half  of  this 
field  of  view  (fig.  5a)  is  the  same  area  as  that 
imaged  in  the  left  half  of  figure  4a.    Some  of 
the  most  highly  attenuating  areas  seen  in  figure 


5a  become  the  least  attenuating  areas  as  the 
sample  is  rotated  in  the  sound  field.    The  acoustic 
interferogram  of  the  same  region  is  shown  in  figure 
5b.    Again,  the  spatial  distribution  character  of 
sonic  velocity  in  the  upper  portion  of  the  micro- 
graph is  similar  to  that  seen  in  figure  4b.  The 
sonic  velocity  in  this  region  is  1580  m/s  with 
typical  variations  less  than  ±  5  m/s.    In  the 
darker  regions  the  interferogram  lines  become 
quite  jagged,  indicative  of  sharply  localized 
changes  in  sonic  velocity.    Typical  shifts  here 
are  on  the  order  of  0.8  normalized  fringes,  cor- 
responding to  an  increase  in  velocity  of  40  m/s. 

Figures  6a  and  6b  show  a  comparison  between 
interferograms  obtained  on  normal  and  infarcted 
tissue.    The  interferogram  obtained  on  the  normal 


76 


(b) 

Fig.  6.    Interferograms  comparing  normal  and  in- 
farcted  tissue,    (a)    Interferogram  of 
normal  tissue  shows  uniform  attenuation 
and  velocity.    Velocity  variations  are 
typically  ±  5  m/s.    (b)    Interferogram  of 
infarcted  tissue.    Both  attenuation  and 
sonic  velocity  are  highly  variable. 
Typically,  the  highly  attenuating  areas 
show  increased  sonic  velocity,  for  example, 
the  region  indicated  by  the  arrow  has  a 
velocity  of  75  m/s  higher  than  the  sur- 
rounding tissue. 

specimen  (fig.  6a)  shows  similar  characteristics 
as  discussed  previously,  i.e.  uniform  attenuation 
and  velocity  over  the  field  of  view.    In  contrast, 
the  most  distinctive  feature  of  the  infarcted  tis- 
sue (fig.  6b)  is  its  overall  microelastic  in- 
homogeneity.    There  are  small  regions  of  this 
micrograph  which  show  uniform  velocity  profiles 
and  attenuation  characteristics  (similar  to  normal 
tissue),  however,  there  also  exists  a  number  of 
highly  attenuating  (darker)  areas.    These  dark 
areas  show  increases  in  sonic  velocity  of  more 
than  75  m/s  compared  to  that  seen  in  the  neighbor- 
ing tissue.    The  complex  interference  fringe  pat- 
tern indicates  that,  in  the  infarcted  zones,  there 
is  considerable  variation  in  the  sonic  velocity 
on  a  microscopic  scale. 

On  several  sections,  ultrasonic  attenuation 
measurements  were  made  at  100  MHz.    The  results 


Table  1.    Summary  of  100  MHz  attenuation 


measurements. 

Section  Attenuation  Number  of 
  range  (cm-i)  measurements 

A  non-infarct              48  l 

B  non-farct^                43b  2 

B  non-infarctc          41-68  2 

C  non-infarct            37-61  2 

D  infarct                  38  -  100  5 

E  infarct                  27-70  5 


Off-axis  component  of  sound  is  perpendicular 
^to  grain. 

Off-axis  component  of  sound  is  parallel  to 
^grain. 
Typical  value. 

are  presented  in  table  1.    Sections  A  through  C 
were  obtained  from  normal  regions  while  sections 
D  and  E  were  obtained  from  zones  of  infarction. 
Each  attenuation  measurement  was  made  on  a  small 
region  of  tissue  (1  x  10~'*cm2)  gp^^  general, 
several  different  areas  on  each  section  were 
measured.    The  range  of  attenuation  measured  in  a 
section  is  given  in  column  2  of  the  table.  The 
variations  tabulated  here  represent  structural 
variations  in  the  tissue  as  opposed  to  uncertain- 
ties in  the  measurement  technique  (which  are  on 
the  order  of  ±  10  percent).    The  number  of  mea- 
surements made  on  each  section  was  governed  pri- 
marily by  the  image  contrast  as  seen  on  the  TV 
monitor.    Areas  that  looked  distinct  on  the  moni- 
tors were  subsequently  measured.    Thus,  sections 
with  uniform  attenuation  properties  were  subjected 
to  fewer  measurements  than  sections  displaying  a 
variety  of  brightness  levels. 

The  attenuation  values  obtained  in  the  3  sec- 
tions of  normal  tissue  range  from  37  to  68  cm"^  . 
For  section  B,  two  entries  are  presented  in  the 
table  corresponding  to  the  two  different  orienta- 
tions of  the  section  in  the  sound  field.  The 
first  entry,  Bi,  represents  measurements  made  on 
section  B  in  the  orientation  depicted  in  figure 
4a,  while  the  second  entry,  B,,,  presents  measure- 
ments made  in  the  orientation  depicted  in  figure 
5a.    The  attenuation  measured  in  the  B  orientation 
show  relatively  minor  variations  from  the  typical 
value  of  43  cm--'.    In  contrast,  measurements  on  the 
same  section  in  the  B,,  orientation  yield  values 
ranging  from  41  to  68  cm-^.    Perhaps  the  most 
striking  feasure  of  the  acoustic  anisotropy  is  the 
observation  that  the  same  area  that  yields  an 
attenuation  coefficient  of  45  cm-^  in  orientation 
Bi  gives  a  value  of  68  cm-^  in  the  B,,  orientation. 
Thus,  variations  in  attenuation  which  are  attrib- 
uted solely  to  anisotropic  propagation  are  of  the 
same  magnitude  as  variations  observed  in  different 
sections . 

The  attenuation  measured  in  the  zones  of  in- 
farction serve  to  quantify  the  visual  impression 
given  by  the  acoustic  micrographs.  Infarcted 
zones  show  inhomogeneous  attenuation  properties 
with  variations  ranging  between  27  and  100  cm"^ 
for  the  two  sections  analyzed.    The  highly  attenu- 
ating zones  with  attenuation  coefficients  in  ex- 
cess of  70  cm"^,  are  morphologically  distinct  from 
any  features  present  in  the  non-infarcted  regions. 


77 


5.  Discussion 

This  study  presents  some  acoustic  microstruc- 
ture  characteristics  of  myocardial  tissue  at  a 
frequency  of  100  MHz.    The  features  revealed  in 
the  acoustic  micrographs  as  well  as  the  quantita- 
tive attenuation  and  velocity  data  can  lead  to 
important  conclusions  with  regard  to  the  complete 
ultrasonic  characterization  of  tissues. 

First  of  all,  100  MHz  acoustic  micrographs  can 
be  used  to  distinguish  between  and  recognize  in- 
farcted  and  normal  tissue.    The  most  distinctive 
feature  of  the  infarcted  zone  is  the  high  degree 
of  localized  variations  in  attenuation  and  veloci- 
ty.   Such  acoustic  contrast  could  result  from 
fibrous  scar  tissue  distributed  in  a  matrix  of 
normal  myocardium.    The  highly  attenuating,  high 
sonic  velocity  regions  can  be  attributed  to  areas 
of  substantial  scar.    This  identification  is  con- 
sistent with  lower  frequency  measurements  on  tis- 
sues rich  in  collagen  [12]. 

Secondly,  significant  acoustic  anisotropy, 
arising  from  muscle  fiber  orientation,  has  been 
observed  at  100  MHz.    One  of  the  sections  analyzed 
(sample  B)  showed  a  large  orientational  dependence, 
and  this  may  have  been  due  to  the  particular  geom- 
etry of  the  fibers.    With  a  small  component  of  the 
propagation  vector  K  along  the  fiber  an  increase 
in  attenuation  was  observed.    This  is  in  the  same 
direction  as  the  attenuation  anisotropy  previously 
observed  in  striated  muscle  at  lower  frequencies 
[13].    The  magnitude  of  change  in  attenuation  is 
rather  substantial,  45  to  68  cm"^,  indicating  that 
structural  aspects  and  tissue  architecture  play  an 
important  role  in  determining  overall  attenuation. 
The  orientation  effect  favors  scattering  type  loss 
mechani  sm. 

Thirdly,  the  large  velocity  variations  measured 
in  the  infarcted  tissue  may  have  important  con- 
sequences for  low  frequency  attenuation  measure- 
ments.   Phase  cancellation  losses  have  been  im- 
plicated as  a  primary  source  of  artifact  in  at- 
tenuation measurement  made  using  piezoelectric 
receivers  [14,15].    Although  small  diameter 
(~  .2  cm)  receiving  transducers  or  confocal  pairs 
of  transducers  can  be  used  to  minimize  phase  can- 
cellation loss,  they  do  not  completely  eliminate 
it.    Phase  cancellation  occurs  when  inhomogenei ties 
in  the  tissue  distort  the  ultrasonic  phase  fronts 
presented  to  a  spatially  extended  piezoelectric 
receiving  transducer.    The  i nterferogram,  figure 
6b,  displays  the  high  degree  of  wavefront  distor- 
tion which  actually  occurs  in  infarcted  myocardium. 
Quantitatively,  the  distortion  is  greater  than  1.5 
fringes  or  540°  in  phase  over  a  distance  which  is 
smaller  than  the  diameter  of  the  small  receiving 
transducer  (.2  cm)  measurement  reported  at  lower 
frequency  [2-4]. 

Lastly,  it  is  interesting  to  compare  the  attenu- 
ation measurements  made  at  100  MHz  with  those  ob- 
tained at  lower  frequencies.    Figure  7  is  a  plot 
of  the  attenuation  coefficient  divided  by  frequency 
(a/f)  vs  f.    The  data  in  the  frequency  range  2  to 
10  MHz  were  reported  previously  and  are  average 
attenuation  values  obtained  on  fresh  myocardium 
excised  4  to  5  weeks  after  vascular  occlusion  [4]. 
On  the  other  hand,  the  100  MHz  data,  taken  from 
table  1,  represent  the  range  of  attenuation  values 
in  the  infarcted  and  non-infarcted  zones. 

Analysis  of  figure  7  indicates  that  for  normal 
tissue,  in  the  frequency  interval  2  to  10  MHz,  a/f 
has  a  relatively  constant  value  of  .07  cm'^  MHz"i. 


non-i  nfarct 


J  ]  I  I  \  L_ 

2  5  10  20  50  100 

Frequency  (MHz) 


Fig.  7.    Comparison  of  100  MHz  attenuation  data 
with  that  obtained  at  lower  frequencies 
for  infarcted  and  non-infarcted  myocardium. 
Low  frequency  data  are  taken  from  the  work 
of  Miller  et  al .  [4]  while  the  100  MHz 
data  are  from  table  1. 

The  100  MHz  values  reported  here  are  substantially 
higher  (.37  to  .68  cm'^  MHz'M-    For  the  infarcted 
tissue,  the  2  to  10  MHz  a/f  values  increase  with 
frequency.    Linear  extrapolation  of  these  data 
would  predict  a  value  of  1.9  cm~^  MHz"^  at  100  MHz. 
However,  the  most  highly  attenuating  regions  ob- 
served in  the  infarcted  zones  at  100  MHz  have  a/f 
values  of  1.0  cm"i  MHz"^,  a  value  significantly 
lower  than  that  predicted.    Thus,  the  a/f  versus 
f  dependence  for  normal  and  infarcted  tissue  pre- 
dicted in  the  range  1  to  10  MHz  may  no  longer  hold 
in  the  interval  10  to  100  MHz. 

The  following  limitations  to  this  conclusion 
should  be  pointed  out.    First  of  all,  the  referred 
to  low  frequency  data  were  obtained  on  freshly  ex- 
cised tissue,  while  this  100  MHz  investigation 
used  formalin  fixed  tissues.    Although  data  ob- 
tained on  other  tissues,  e.g.  ,  kidney,  indicate 
that  the  effect  of  formalin  on  the  overall  attenu- 
ation values  at  100  MHz  may  not  be  significant 
[10],  this  has  not  yet  been  verified  for  the  myo- 
cardium.   The  second  point  concerns  orientation 
effects.    The  lower  frequency  measurements  were 
made  by  transmitting  sound  through  the  myocardial 
wall  roughly  perpendicular  to  the  epicardial  sur- 
face, while  the  100  MHz  measurements  were  done  in 
a  plane  perpendicular  to  this.    Results  presented 
in  this  study  indicate  the  presence  of  an  orienta- 
tional effect  on  the  acoustic  properties. 

Although  more  comprehensive  work  needs  to  be 
done  to  quantify  the  extent  to  which  orientation 
and  fixing  influence  the  acoustic  properties,  the 
results  of  this  study  show  that  high  frequency 
acoustic  microscopy  can  play  an  important  role 
in  the  understanding  and  characterization  of  tis- 
sues at  diagnostic  frequencies. 

Acknowledgment 

The  stimulating  discussions  with  W.  D.  O'Brien, 
University  of  Illinois  are  gratefully  acknowledged. 

References 

[1]    Lele,  P.  P.  and  Namery,  J.,  A  computer-based 
ultrasonic  system  for  detection  and  mapping 


78 


of  myocardial  infarcts,  in  Proc.  San  Diego  [9] 
Biomed.  Symp.  U,  121  (19721: 

[2]    Yuhas,  D.  E.,  Mimbs,  J.  W. ,  Miller,  J.  G., 
Weiss,  A.  N.,  and  Sobel ,  B.  E.,  Changes  in 
Ultrasonic  Attenuation  Indicative  of  Regional  [10] 
Myocardial  Infarc'-'on,  In  Ultrasound  in 
Medicine ,  D.  White,  ed..  Vol .  3  (Plenurfi'  Press , 
New  York,  1977). 

[11] 

[3]    Mimbs,  J.  W.,  Yuhas,  D.  E.,  Miller,  J.  G., 
Weiss,  A.  N.,  and  Sobel,  B.  E.,  Detection  of 
myocardial  infarction  in  vitro  based  on 
altered  attenuation  of  ultrasound.  Circulation 
Research  (in  press). 

[4]    Miller,  J.  G.,  Yuhas,  D.  E.,  Mimbs,  J.  W.,  [12] 
Dierken,  S.  B.,  Busse,  L.  J.,  Weiss,  A.  N., 
and  Sobel,  B.  E.,  Ultrasonic  Tissue  Charac- 
terization:   Correlation  between  Biochemical 
and  Ultrasonic  Indices  of  Myocardial  Injury, 
in  Proc.  of  IEEE  Ultrasonic  Symposium,  J. 
deKlerk  and  B.  McAvoy,  eds..  Cat.  No.  76  [13] 
Ch.  1120-4SU,  p.  33  (Annapolis,  1976). 

[5]    Korpel ,  A.,  Kessler,  L.  W.,  and  Palermo, 
P.  R.,  An  acoustic  microscope  operating 
at  100  MHz,  Nature  232,  110  (1971). 

[6]    Kessler,  L.  W. ,  Korpel,  A.,  and  Palermo, 
P.  R. ,  Simultaneous  acoustic  and  optical 
microscopy  of  biological  specimens. 
Nature  239,  111  (1972).  [14] 

[7]    Kessler,  L.  W. ,  Palermo,  P.  R. ,  and  Korpel, 
A.,  Practical  High  Resolution  Acoustic 
Microscopy,  in  Acoustical  Holography,  G.  [15] 
Wade,  ed.  ,  Vol.  4,  p.  51  (Plenum  Press, 
New  York,  1972). 

[8]    Kessler,  L.  W.,  Palermo,  P.  R.,  and  Korpel, 
A.,  Recent  Developments  with  the  Scanning 
Laser  Acoustic  Microscope,  in  Acoustical 
Holography,  P.  S.  Green,  ed..  Vol.  5,  p.  15 
(Plenum  Press ,  New  York,  1974). 


Kessler,  L.  W.,  The  Sonosmicroscope ,  in  Proc. 
of  IEEE  Ultrasonic  Symposium,  J.  deKlerk, 
ed.,  p.  735,  Cat.  No.  74-CH0-896-1  SU  (IEEE, 
New  York,  1974). 

Kessler,  L.  W. ,  VHP  ultrasonic  attenuation 
in  mammalian  tissue,  J.  Acoust.  See.  Am.  53 
1759  (1973). 

Kessler,  L.  W.,  Tissue  Characterization  by 
Means  of  Acoustic  Microscopy,  in  Ultrasonic 
Tissue  Characterization,  M.  Linzer,  ed.. 
Spec.  Publ .  453,  p.  261  (U.S.  Government 
Printing  Office,  Washington,  D.C.,  1976). 


O'Brien,  W.  D.,  Jr.,  The  Role  of  Collagen  in 
Determining  Ultrasonic  Propagation  Proper- 
ties in  Tissue,  in  Acoustical  Holography, 
L.  W.  Kessler,  ed. ,  Vol.  7  (Plenum  Press, 
New  York,  1977). 

Dussik,  K.  T.  and  Fritch,  D.  J.,  Determina- 
tion of  Sound  Attenuation  and  Sound  Velocity 
in  the  Structures  Constituting  the  Joints 
and  of  the  Ultrasonic  Field  Distribution 
within  the  Joints  on  Living  Tissues  and 
Anatomical  Preparations,  both  in  Normal  and 
Pathological  Condition,  Progress  Report, 
Project  A454,  Public  Health  Service,  April 
(1955),  September  (1956). 

Marcus,  P.  N.  and  Carstensen,  E.  L.,  Problems 
with  absorption  methods  of  inhomogeneous 
solids,  J.  Acoust.  Soc.  Am.  58,  1334  (1975). 

Busse,  L.  J.,  Miller,  J.  G.,  Yuhas,  D.  E., 
Mimbs,  J.  W.,  Weiss,  A.  N.,  and  Sobel,  B.  E., 
Phase  Cancellation  Effects:    A  Source  of  At- 
tenuation Artifact  Eliminated  by  a  CdS 
Acoustoelectric  Receiver,  in  Ultrasound  in 
Medicine,  D.  White,  ed..  Vol.  3,  pp.  1519- 
1535"TPTenum  Press,  New  York,  1977). 


79 


i  5 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ACOUSTIC  PROPERTIES  OF  NORMAL  AND  ABNORMAL  HUMAN  BRAIN 


F.  W.  Kremkau,  C.  P.  McGraw,  and  R.  W.  Barnes 

Bowman  Gray  School  of  Medicine 
Winston-Salem,  North  Carolina    27103,  U.S.A. 


Attenuation  and  propagation  speed  at  1,  3,  and  5  MHz  and  specific  acoustic  im- 
pedance at  2  MHz  were  measured  j_n  vitro  in  10  tissue  samples  from  8  abnormal  human 
brains.    The  results  were  compared  to  values  resulting  from  a  study  of  5  normal 
human  brains  reported  elsewhere.    Conclusions  from  literature  review  are  that  attenua- 
tion is  lower  in  gliobastoma  than  in  normal  brain  and  attenuation,  propagation  speed 
and  impedance  are  higher  in  meningioma  than  in  normal  brain.    Conclusions  from  the 
data  of  the  present  study  are  that  subarachnoid  hemorrhage  appears  acoustically 
normal;  infarcts  have  normal  attenuation  but  increased  speed;  clots,  intracerebral 
hemorrhage,  and  metastases  have  increased  attenuation  and  speed;  hydrocephalic  brain 
has  very  low  attenuation  and  low  speed.    In  general,  it  appears  that  abnormal  condi- 
tions produce  increases  in  all  three  acoustic  properties  studied. 

Key  words:    Attenuation;  brain  tumor;  clot;  hemorrhage;  hydrocephalus;  impedance; 
infarct;  speed;  ultrasonic. 


1.  Introduction 

As  part  of  a  study  of  acoustic  properties  of 
normal  and  abnormal  tissues,  attenuation  and 
propagation  speed  at  1,  3,  and  5  MHz  and  specific 
acoustic  impedance  (hereafter  called  impedance) 
at  2  MHz  were  measured  in  vitro  in  10  tissue 
samples  from  8  abnormal  human  brains. 

Results  from  our  study  of  23  tissue  samples 
from  5  normal  human  brains  are  reported  else- 
where [l]i.    The  important  observations  were: 

(a)  attenuation  measured  with  piezoelectric 
transducers  resulted  in  higher  values 
(0  to  50  percent)  than  those  obtained 
using  radiation  force  method; 

(b)  attenuation  (using  transducer  method) 
was  0.9  dB/cm  at  1  MHz  and  was  a  function 
of  f^"-'  over  the  range  1  to  5  MHz; 

(c)  propagation  speeds  at  1  MHz  were  1546 
and  1539  m/s  for  fresh  and  fixed  tissues, 
respecti  vely ; 

(d)  speed  dispersion  was  2  ms"^  MHz-^  and 

1.6  ms"i  MHz-i  for  fresh  and  fixed  tissues, 
respectively; 

(e)  white  matter  had  attenuation  1.4  times 
that  for  gray; 

(f)  attenuation  of  adult  brain  was  2.7  times 
that  for  infant; 

(g)  one  day  aging  reduced  attenuation  up  to 
20  percent; 

(h)  attenuation  depended  upon  temperature  to 
the  power  -0.1  and  -0.5  at  1  and  5  MHz, 
respecti  vely ; 

(i)  propagation  speed  as  a  function  of  tem- 
perature exhibited  a  minimum  at  approxi- 
mately 15  °C. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


The  differences  between  white  and  gray  matter  and 
between  infant  and  adult  attenuation  can  be  ex- 
plained on  the  basis  of  tissue  water  content. 
The  minimum  in  propagation  speed  with  respect  to 
temperature  may  indicate  a  sensitivity  to  the 
crystal -1 iquid  crystal  phase  transition  for  mem- 
brane phospholipid  bilayers. 

There  are  several  reports  in  the  literature 
regarding  the  acoustic  properties  of  normal  and 
abnormal  human  brain  tissue.    A  review  of  the 
data  for  normal  brain  may  be  found  in  our  earlier 
report  [1].    In  addition,  reported  values  for  im- 
pedance are  1.59  x  10^  MKS  rayl  (kg/m^s)  for 
"brain"  [2],  1.60  for  cerebellum  [2],  1.51  for 
cerebrum  [3],  and  1.52  for  white  matter  [4]. 

An  early  paper  by  Wild  and  Reid  [5]  showed 
differences  in  received  echo  patterns  for  normal 
brain  and  several  malignant  and  benign  tumors. 
More  recently  Fishman.  Heyser,  and  Le  Croissette 
[6]  reported  high  attenuation  in  tumor  regions  of 
human  brain  sections  using  qualitative  transmis- 
sion images.    A  summary  of  the  data  for  abnormal 
brain  is  given  in  table  1.    Only  tumors  are  cited 
as  no  other  abnormality  was  found  in  the  litera- 
ture.   Table  2  shows  an  attempt  to  simplify  this 
summary  and  draw  some  conclusions  from  it.  It 
lists  reports  of  greater  (+),  equal  (==),  or  lesser 
(-)  values  for  attenuation,  speed,  or  impedance 
as  compared  to  the  same  authors'  values  for  normal 
brain  tissue.    Citing  only  those  situations  where 
at  least  two  reports  are  in  agreement,  the  follow- 
ing are  concluded: 

(a)  attenuation  is  lower  in  glioblastoma  than 
in  normal  brain; 

(b)  attenuation  is  higher  in  meningioma  than 
in  normal  brain; 

(c)  propagation  speed  and  impedance  are 
higher  in  meningioma  than  in  normal  brain. 


81 


Table  1.    Attenuation,  propagation  speed,  and  impedance 
for  several  brain  tumors--l iterature  summary. 

Tissue    Frequency  Attenuation       Speed      Impedance  Reference 
(MHz)         (dB/cm)  (m/s)  {^0''  MKS  rayl ) 


2 
7 

2 

7 
8 
9 
9 

7 

2 
7 
9 

7 
10 
11 
12 
13 

2 
9 
9 


2 
7 
3 
9 
9 
10 
12 
13 
14 

3 
12 

2 
7 
10 
4 

7 

7 

7 
3 
2 

apormalin  fixed  sampl es . 


2.    Materials  and  Methods 

Normal  and  abnormal  human  brain  tissue  samples 
were  obtained  at  autopsy.    Measurements  were  made 
at  37  °C  after  at  least  24  hours  of  fixation  in 
10  percent  formalin.    Except  during  measurement, 
tissues  were  stored  at  4  °C.    Tissue  samples  were 
cut  to  fill  the  sample  holder  -  a  Teflon®  ring  with 
inner  diameter  33  mm,  outer  diameter  45  mm,  and 
(sample)  length  15  mm.    Saran®  membranes  were 
stretched  over  both  sides  of  the  chamber  to  pro- 
vide flat  tissue  surfaces  and  acoustic  windows. 
The  sample  holder  fit  into  a  guide  in  a  tempera- 
ture controlled  water  bath  which  provided  con- 
sistent placement  (within  1  mm  in  each  coordinate) 
relative  to  source  and  receiver  transducers.  The 
transducers  were  one  inch  diameter  Val pey-Fi sher 
polished  1  MHz  X-cut  quartz.    The  source  trans- 
ducer was  driven  by  an  Arenberg  PG-650C  pulsed 
oscillator.    Bursts  were  typically  20  cycles  long 
and  reasonably  characterized  by  a  single  (funda- 
mental) frequency.    The  receiver  transducer  was 
connected  to  the  input  of  a  7A18  amplifier  in  a 
Tektronix  7904  oscilloscope  with  a  7B92  dual  time 
base.    Amplitude  of  received  burst,  with  and 
without  tissue  sample  between  transducers,  was 
measured  and  attenuation  coefficient,  calculated 
from  the  following  relation: 


Table  2.    Attenuation,  propagation  speed,  and  impedance  are  higher 
than  (+),  equal  to  (=),  or  lower  than  (-)  that  for  normal 
tissue  as  measured  by  the  same  authors.    There  is  a  con- 
sensus that  glioblastoma  has  lower  attenuation  and  that 
meningioma  has  higher  attenuation,  speed,  and  impedance. 


Tumor  Attenuation 

Speed 

Impedance  Reference 

acoustic  neurinoma 

+ 

+ 

2 
7 

arachnoidal  sarcoma 

2 

astrocytoma 

+ 

+ 

7 
8 

craniopharyngioma  + 

7 

ependymoma 

+ 

+ 

+ 

2 
7 

glioblastoma 

+ 

7 

10 
11 

12 
13 

g 1 i  oma 

2 

medul loblastoma  + 

7 

meningioma 

+ 

+ 
+ 
+ 
+ 

+ 
+ 

+ 
+ 

2 
7 
3 
10 
12 
1  3 
14 

metastatic  adenocarcinoma 

- 

+ 

+ 

3 
12 

metastatic  carcinoma 

+ 

+ 

+ 

2 
7 
10 
4 

oligodendroglioma 

7 

pineal oma 

7 

pituitary  adenoma 

7 

retinoblastoma 

+ 

■ 

3 

spongioblastoma 

2 

m  =  ^   20  log^o 

Ai  " 

where  m  and  ni,^  are  attenuation  coefficients  in 
dB/cm  of  tissue  sample  and  water,  respectively, 
m^  is  window  loss  in  dB,  d  is  sample  thickness  in 
cm,  Aq  is  received  amplitude  with  sample  not 
present  (water  path),  and  Aj  is  received  amplitude 
with  sample  present.    Amplitudes  were  measured 
with  precision  ±  2  percent  or  better  and  accuracy 
(based  on  Tektronix  specification)  within  2  per- 
cent.   Measurements  were  taken  at  1,  3,  and  5 
MHz.    Corrections  for  m^  and  m^  are  normally  less 
than  1  percent  except  m^,  at  3  and  5  MHz  (typical- 
ly 2  and  7  percent,  respectively).    The  value  for 
m^-  is  obtained  by  measuring  the  received  ampli- 
tude change  when  a  water  filled  chamber  is  in- 
serted between  the  transducers.    Additional  re- 
flection loss  resulting  from  impedance  discon- 
tinuity between  water  and  tissue  is  not  signifi- 
cant. 

Propagation  speed  was  calculated  using  the  ar- 
rival time  change  occurring  when  the  sample  was 
inserted  between  source  and  receiver  transducers, 
c  =  d/(t2  -  t  )  where  c  is  propagation  speed  in  the 
tissue  sample^in  m/s,  d  is  sample  thickness  in 
meters,  tj  is  propagation  time  in  seconds  in  d  path 
length  of  water,  and  t    is  arrival  time  change  in 
seconds  when  sample  is  introduced  into  sound  path. 
Arrival  time  changes  are  typically  200  to  400  ns 
read  to  a  precision  of  +  10  ns  and  accuracy  (based 
on  Tektronix  specification)  within  5  percent. 
Measurements  were  taken  at  1,  3,  and  5  MHz. 


acoustic 
neurinoma 

arachnoidal 
sarcoma 

astrocytoma 


craniopharyngioma 
ependymoma 

gl  ioblastoma 


6-8 


gl ioma 

medulloblastoma 
men i  ng ioma 


4 

5 

2 
2 

5 

1 
1 
5 

metastatic 

adenocarcinoma  1 

metastatic  4 
carcinoma  5 
5 
5 

oligodendroglioma  5 
pineal oma  5 
pituitary  adenoma  5 
retinoblastoma 
spongioblastoma  4 


9-12 

9 

6-9 
3.3a 
20a 
0.5 
0.5 


8-11 
10-13 


6.3a 
0.9-1.2 
0.6-1.2 
35 

0.3 

8-10 
4.3a 

7 
9 
7 


1522-1547  1 
1530 


1660 
1545 
1517a 


1537-1545  1 
1501 


1525-1547  1 
1529 
1500 


1540-1550  1 

1640 
1546-1569 
1524a 


1590 
1535 


1600 
1532 


57-1 .62 
1.58 

1.71 
1  .64a 

60-1 .62 
1  .54 


.57-1 .61 
1  .56 
1  .54 


61-1 .64 
1 .72 
1  .57a 


1.68 
1  .60 

1  .52 


1  .67 
1.59 


82 


Acoustic  impedance  was  measured  using  the 
method  of  Gregg  and  Palagallo  [4].    A  Parametrics 
Pulser/Receiver  5050PR  was  used  with  a  Hoffrel  310A 
2.0  MHz,  13  mm  diameter  diagnostic  transducer. 
Display  was  produced  on  the  Tektronix  7904  oscil- 
loscope. 

3.  Results 

The  results  of  seven  abnormal  conditions  are 
given  in  table  3.    With  respect  to  attenuation 
and  propagation  speed,  the  following  observations 
are  made: 

(a)  subarachnoid  hemorrhages  appear  normal 
acoustically; 

(b)  infarcts  have  normal  attenuation  but 
increased  speed; 

(c)  clots,  intracerebral  hemorrhage,  and 
metastases  have  increased  attenuation 
and  speed; 

(d)  hydrocephalic  brain  has  very  low  at- 
tenuation and  low  propagation  speed. 

Impedance  values  do  not  correlate  well  with  tis- 
sue type  or  with  propagation  speed  values. 


Table  3.    Acourtic  properties  for  seven  abnormal  conditions 
in  human  brain. 


Tissue 

Attenuation 
(dB/cm) 

Speed 
(m/s) 

Impedance 
(10''  MK5  rayl  ) 

1  MHz 

3  MHz 

5  MHz 

1  MHz 

2  MHz 

normals 

0.9 

2.8 

5.2 

1539 

1  .43 

metastatic  lung 
carcinoma 

1.5 

5.6 

11.0 

1568 

1  .56 

ventricular 
clotb 

2.3 

9.6 

12.2 

1575 

1  .55 

subarachnoid 
hemorrhage^ 

0.9 

3.2 

5.6 

1535 

1  .73 

intracerebral 
hemorrhage 

1.9 

8.2 

12.8 

1553 

1  .78 

hemorrhagic  infarct 
with  necrosis 

0.9 

3.0 

5.7 

1552 

1.72 

Island  of  Kiel 
infarct 

0.9 

3.2 

6.6 

1556 

1  .46 

hydrocephalus 

0.2 

0.8 

1.3 

1516 

2.13 

^See  reference  [1].  Impedance  value  was  obtained  by  measurement 
on  one  normal  brain  of  the  five  studied  with  respect  to  attenu- 

j^ation  and  propagation  speed. 
Mean  of  three  samples. 
Mean  of  two  samples. 


4.  Discussion 

In  general,  it  appears  that  abnormal  condi- 
tions produce  increases  in  all  three  acoustic 
properties  studied.    The  measured  metastatic 
carcinoma  values  agree  reasonably  well  with 
those  of  others  [4,7]  (see  table  1).    The  low 
attenuation  for  hydrocephalus  is  similar  to 
that  reported  for  infant  brain  where  high  water 
content  was  suggested  as  an  explanation  [1]. 
The  low  propagation  speed  for  hydrocephalus  was 
not  observed  in  the  normal  infant.    Our  results 
show  normal  attenuation  for  infarcts  (1  to  5 
MHz).    Mi  1 ler  et  al .  [15]  observed  an  increase 
in  attenuation  for  dog  heart  after  infarct  but 
this  was  evidenced  only  at  frequencies  above 
5  MHz.    Lele  and  Namery  [16]  reported  increased 
attenuation  above  2  MHz  for  infarcted  dog  heart 
compared  to  normal.    They  also  observed  reduced 
impedance  for  infarct  which  was  not  observed  in 
our  study  of  brain. 


Few  of  these  abnormal  tissues  have  been 
studied  by  more  than  one  group.    Much  more  data 
will  be  required  before  confident  generaliza- 
tions can  be  agreed  upon. 

Acknowledgments 

The  authors  are  grateful  for  technical  assist- 
ance from  Ms.  S.  Gaskey. 

This  work  was  supported  in  part  by  NINCDS 
Contract  NOl-NS-4-2304. 


References 

[1]    Kremkau,  F.  W.,  McGraw,  C.  P.,  and  Barnes, 

R.  W. ,  Ultrasonic  attenuation  and  preparation 
speed  in  normal  human  brain,  submitted  for 
publication,  J.  Acoust.  Soc.  Am. 

[2]  Schiefer,  W. ,  Kazner,  E. ,  and  Kunze,  S. , 
Clinical  Echo-Encephalography ,  pp.  67-68 
(Springer-Verlag,  New  York,  1968 ) . 

[3]    Ishikawa,  S.,  Yukishita,  K. ,  and  Ito,  K. , 

Ultrasonic  attenuation  in  brain  (7th  report), 
Jap.  Med.  Ultrasonics  3,  33  (1965),  quoted 
in  reference  7. 

[4]    Gregg,  E.  C.  and  Palagallo,  G.  L.,  Acoustic 
impedance  of  tissue.  Investigative  Radiol .  4, 
357-363  (1969). 

[5]    Wild,  J.  J.  and  Reid,  J.  M. ,  The  effects  of 
biological  tissues  on  15-mc  pulsed  ultra- 
sound, J.  Acoust.  Soc.  Am.  25^,  270-280  (1953). 

[6]    Fishman,  L.  S.  ,  Heyser,  R.  C,  and  Le 

Croissette,  D.  H.,  Ultrasonic  transmission 
measurements  in  human  brain  sections. 
Radiology  112,  211-213  (1974). 

[7]    Ishikawa,  S.,  Yukishita,  K. ,  Sato,  K. ,  Ito, 
K.  ,  and  Wagai,  T.  ,  Ultrasonic  attenuation  in 
brain  tissue  (the  7th  report),  Jap.  Med. 
Ultrasonics  3,  48  (1965). 

[8]    Uematsu,  S.  and  Walker,  A.  E.,  A  Manual  of 
Echoencephalography ,  p.  40  (Williams  and 
Wilkins,  Baltimore,  1971). 

[9]    Van  Venrooij,  G.  E.  P.  M.,  Measurement  of 
ultrasound  velocity  in  tissue.  Ultrasonics 
9,  240-242  (1971). 

[10]    Tanaka,  K. ,  Yukishita,  K. ,  Ito,  K. ,  Ehara, 
K. ,  and  Watanabe,  H. ,  Ultrasonic  diagnosis 
of  brain  tumors,  Proc.  Intl.  Symp.  on  Echo- 
Encephalography,  (1967),  pp.  38-44,  quoted 
in  Erikson,  K.  R. ,  Fry,  J.  J.,  and  Jones, 
J.  P.,  Ultrasound  in  medicine  -  a  review, 
IEEE  Trans.  Sonics  Ultrasonics  SU-21,  144- 
170  (1974). 

[11]    Le  Croissette,  D.  H.  and  Heyser,  R.  D. , 
Attenuation  and  Velocity  Measurements  in 
Tissue  Using  Time  Delay  Spectrometry,  in 
Ultrasonic  tissue  Characterization,  M. 
Linzer,  ed..  National  Bureau  of  Standards 
Spec.  Publ .  453,  pp.  81-95  (U.S.  Government 
Printing  Office,  Washington,  D.C.,  1976). 


83 


[12]    Oka,  M.  and  Yosioko,  K. ,  Ultrasonic  absorp- 
tion of  human  brain  tissue.  Paper  #1302 
presented  at  World  Congress  of  Ultrasound 
in  Medicine,  San  Francisco  (1976). 

[13]    Nakaima,  N.,  Study  of  the  ultrasonic  at- 
tenuation of  brains  bearing  tumors,  J^. 
Wakayama  Med.  Soc.  27,  69-106  (1976). 

[14]    Kikuchi,  Y.,  Tanaka,  K. ,  and  Wagai,  T. , 

Early  cancer  diagnosis  through  ultrasonics, 
J.  Acoust.  Soc.  Am.  29,  824-833  (1957). 

[15]    Miller,  J.  G.,  Yuhas,  D.  E.,  Mimbs,  J.  W., 
Dierker,  S.  B.,  Busse,  L.  J.,  Laterra, 
J.  J.,  Weiss,  A.  N.,  and  Sobel  ,  B.  E., 
Ultrasonic  tissue  characterization:  cor- 
relation between  biochemical  and  ultra- 
sonic indices  of  myocardial  injury,  IEEE 
Ultrasonics  Symposium  Proceedings  (1976) 
IEEE  Catalogue  No.  76  CH  1120-5SU,  pp.  33-43. 


[16]    Lele,  P.  P.  and  Namery,  J.,  A  Computer-Based 
Ultrasonic  System  for  the  Detection  and 
Mapping  of  Myocardial  Infarcts,  in  Proceed- 
ings San  Diego  Biomedical  Symposium,  pp. 
121-132,  San  Diego  Biomedical  Symposium, 
San  Diego,  California,  1974. 


84 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


FREQUENCY  DEPENDENT  ATTENUATION  OF  MALIGNANT  BREAST  TUMORS 
STUDIED  BY  THE  FAST  FOURIER  TRANSFORM  TECHNIQUE 


E.  Kelly  Fry,  N.  T.  Sanghvi,  and  F.  J.  Fry 

Indiana  University  School  of  Medicine  and 
Indianapolis  Center  for  Advanced  Research 
Indianapolis,  Indiana    46202,  U.S.A. 

and 

H.  S.  Gallager 

The  University  of  Texas  System/Cancer  Center 
Houston,  Texas    77025,  U.S.A. 


The  work  discussed  in  this  paper  comprises  one  aspect  of  an  experimental  design 
concerned  with  the  use  of  multi-discipline  examination  methods  to  provide  detailed 
information  on  the  interaction  of  normal  and  malignant  breast  tissues  with  a  high 
frequency  sound  field.    The  complete  experimental  design  included  x-ray  and  needle 
biopsy  examination  of  the  breast  of  a  patient  prior  to  mastectomy,  followed  by:  x-ray 
examination  of  the  excised  breast  (the  malignant  tumor  remained  intact  in  the  excised 
breast);  ultrasound  visualization  and  FFT  studies  of  the  formalin-fixed,  excised 
breast  specimen;  x-ray  examination  of  0.5  cm  thick,  whole  breast  sections  of  the  tis- 
sue; and,  finally,  sectioning  and  histological  staining  of  the  primary  mal ignant 
tumor  region  and  other  tissue  areas  of  interest.    Emphasis  is  given  in  this  prelimi- 
nary report  to  a  study  of  attenuation  of  the  sound  beam  as  a  function  of  frequency 
for  specific  tissue  paths  (i.e.,  from  skin  to  back  surface  of  the  excised  breast), 
which  included  (1)  the  malignant  tumor,  (2)  the  nipple  and  (3)  the  areola. 

For  the  tissue  path  which  included  the  malignant  tumor,  the  FFT  studies  indicate 
that  the  attenuation  for  the  full  range  of  frequencies  studied  (1.1  to  4.4  MHz)  was 
greater  than  that  of  any  other  area  of  the  breast.  A  significant  result  of  the  in- 
vestigations reported  in  this  paper  is  the  determination  that  this  analytical  tech- 
nique is  feasible  and  can  yield  data  on  malignant  and  normal  regions  of  breast  tis- 
sue which  correlate  with  the  ultrasound  visualization  imaging  information  and  with 
the  tissue  structure  information  as  revealed  by  histological  examination. 


Key  words:    Attenuation  of  areola;  attenuation  of  breast  tissue;  attenuation  of 
malignant  tumors;  breast  cancer  detection;  breast  carcinoma;  breast 
examination  techniques;  FFT  techniques  for  breast;  histology  of 
breast  tumors;  signal  processing  for  tissue;  x-ray  examination  of 
breast. 


1.  Introduction 

At  the  present  time,  ultrasound  visualization 
techniques,  if  used  in  conjunction  with  other 
clinical  examination  methods,  can  improve  the 
level  of  success  in  the  differential  diagnosis 
of  breast  pathologies  [1]^.    However,  the  theo- 
retical capability  of  ultrasound  for  providing  a 
successful  method  of  early  detection  of  breast 
cancer  is  not  presently  realized,  in  part  because 
of  the  lack  of  sufficient  clinical  studies  with 
this  method  and,  in  addition,  because  of  the  need 
for  more  basic  experimental  studies  on  the  inter- 
action of  ultrasound  and  breast  tissue.  There 
is,  in  fact,  a  serious  need  for  relatively  so- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


phisticated  investigations  concerned  with  (1) 
the  complex  structure  of  normal  breast  in  terms 
of  its  interaction  with  sound  fields  and  (2)  the 
variability  in  structural  features  of  malignant 
tumors  (including  those  in  the  same  pathology 
classification)  insofar  as  this  variability  is 
significant  to  medical  diagnostic  data  obtained 
by  ultrasound  examination  of  such  tumors. 

The  experimental  work  discussed  in  this  paper 
comprises  one  aspect  of  an  overall  approach  of 
using  multi-discipline  examination  methods,  name- 
ly, x-rays,  ultrasound  visualization,  signal 
processing  of  ultrasound  transmission  data  (i.e., 
Fast  Fourier  Transform  (FFT)  techniques)  and 
histological  techniques  to  provide  detailed  in- 
formation on  the  structural  features  of  breast 
tissue  and  their  interaction  with  a  high  frequency 
sound  field.    In  the  initial  experimental  stages, 


85 


procedures  such  as  the  FFT  are  best  performed  on 
an  excised  breast.    A  special  preparation,  namely, 
an  excised,  formalin-fixed  breast  with  a  known  in- 
tact malignant  tumor,  was  used  in  the  investiga- 
tions discussed  in  this  paper.    Detailed  examina- 
tion of  such  breasts,  by  both  standard  and  ex- 
perimental breast  cancer  detection  methods,  and 
the  correlation  of  the  results  with  tissue  struc- 
ture information  provided  by  histological  in- 
vestigations, can  yield  data  that  is  pertinent 
to  early  breast  cancer  detection  [2-8].  The 
rationale,  validity  and  value  of  studying  such 
fixed,  whole  specimens  by  ultrasound  visualization 
techniques  have  been  discussed  in  detail  by  Kelly 
Fry  and  Gallager  in  a  previous  publication  [8]. 
Additionally,  Calderon  et  al .  have  published  re- 
sults of  an  investigation  on  the  ultrasound  at- 
tenuation values  of  formalin-fixed  breast  tumors 
[9]. 

In  capsule  form,  the  overall  experimental  de- 
sign called  for  collection  of  data  first  by  x-ray 
examination  of  the  breast  of  the  patient  prior  to 
surgery,  followed  by:    x-ray  examination  of  the 
excised  breast;  detailed  ultrasound  visualization 
and  FFT  studies  of  the  formalin-fixed  excised 
breast  specimen;  whole  breast  sectioning  of  the 
tissue  in  blocks  of  approximately  0.5  cm  in  thick- 
ness; x-ray  examination  of  each  of  these  cross 
sections  and,  finally,  sectioning  and  histological 
staining  of  the  primary  malignant  tumor  region  and 
other  tissue  areas  of  interest. 

The  FFT  studies  were  considered  in  the 
nature  of  a  feasibility  study  and  it  is  in  that 
regard  they  are  reported  in  this  paper.  Emphasis 
is  given  in  this  preliminary  report  to  a  study  of 
attenuation  of  the  sound  beam  both  in  the  region 
of  the  malignant  tumor  and  in  other  areas  of  the 
breast  tissue  (such  as  nipple  and  areola)  as  a 
function  of  frequency.    As  used  in  this  paper, 
the  term  "attenuation"  designates  all  losses  in 
sound  pressure  amplitude  or  intensity  as  the 
acoustic  beam  traverses  the  tissue,  including 
those  due  to  specular  reflection,  scattering,  re- 
fraction, absorption  and  diffraction. 

Both  the  location  (as  revealed  by  x-ray  examina- 
tion of  the  whole,  excised  breast)  and  the  malig- 
nant character  of  the  primary  mass  (as  revealed  by 
needle  biopsy)  were  known  at  the  time  of  the  ultra- 
sound visualization  and  FFT  studies.    At  the  time 
of  this  writing,  the  histological  classification 
of  the  primary  malignant  tumor  has  been  determined 
and  detailed  histological  studies  are  underway. 
The  precise  correlations  between  information  re- 
vealed by  histology,  ultrasound  visualization  and 
FFT  techniques  will  be  the  subject  of  a  later 
paper. 

2.    Experimental  Methods 

The  tissue  specinen  used  in  the  subject  study 
was  an  excised,  formalin-fixed,  left  breast  of  a 
female  subject  who  was  49  years  of  age  at  the  time 
of  the  mastectomy.    Since  the  diagnosis  of  breast 
carcinoma  had  been  made  by  x-ray  and  needle  biopsy 
prior  to  surgery,  the  mastectomy  was  carried  out 
without  a  surgical  biopsy  so  that  the  excised 
breast  contained  the  previously  detected  malignant 
mass  in  an  essentially  undisturbed  state.  After 
excision,  the  breast  was  made  to  assume  its  ap- 
proximate normal  contour  by  pinning  it  to  a  layer 
of  paraffin,  was  x-rayed  and  then  formalin-fixed. 
The  fixation  process  consisted  of  covering  the 


pinned  breast  with  10  percent  formalin  solution, 
completely  draining  this  solution  and  adding  a 
fresh  formalin  solution  every  other  day  for  a 
total  period  of  two  weeks.    After  the  completion 
of  this  process,  the  breast  was  maintained  in 
formalin  solution  except  for  periods  of  ultrasonic 
examination. 

The  instrumentation  used  for  pulse-echo  examina- 
tion of  the  excised  breast  was  a  linear  scan,  B- 
mode  visualization  system  which  included  a  variety 
of  types  of  transducers.    Only  two  of  these  trans- 
ducers were  used  as  transmitters  in  the  FFT 
studies,  namely,  1.1  MHz  and  4.4  MHz  center  fre- 
quency units.    Most  of  the  FFT  studies  were  car- 
ried out  with  the  4.4  MHz  unit  and  the  results  ob- 
tained with  this  transducer  will  be  discussed  in 
this  paper.    This  transducer  has  a  diameter  of 
1.9  cm,  a  7.5  cm  focal  length,  a  midband  frequency 
response  of  4.4  MHz,  a  band  width  of  3  MHz  jutt 
above  noise,  and  1  MHz  at  the  3  dB  point. 

During  the  visualization  scanning  procedures, 
the  breast  and  examining  transducers  were  com- 
pletely immersed  in  mammalian  Ringer's  solution 
at  room  temperature  (23  °C)  so  that  there  was 
direct  fluid  coupling  of  the  ultrasound  to  the 
tissue.    Both  transverse  (medial-to-lateral)  and 
longitudinal  (superior-to-inferior)  scans  were 
taken  at  distance  intervals  of  1  to  2  mm.  In 
order  to  relate  data  recorded  on  the  echograms  to 
specific  areas  of  the  tissue,  an  anatomical  land- 
mark on  the  tissue  (such  as  the  center  of  the 
nipple  or  the  center  of  a  prominent  skin  discolor- 
ation over  the  area  of  abnormality)  was  selected 
and  a  highly  reflective  and  attenuating  acoustic 
target  placed  on  this  landmark.    With  the  focus 
of  the  transducer  set  on  the  target  center,  the 
echogram  of  the  breast  tissue  showed  a  distinct, 
easily  identified  front  surface  reflection  and  an 
attenuation  shadow  of  the  target.    Since  the 
linear  coordinates  (X  and  Y)  were  recorded  for 
each  echogram,  subsequent  scans  of  the  tissue  with 
the  target  removed  could  be  directly  related  to 
the  chosen  anatomical  landmark.    These  surface 
landmarks  were  subsequently  used  in  the  FFT  ex- 
periments for  identification  of  desired  sound 
transmission  tissue  paths. 

For  the  ultrasound  attenuation  studies  in  the 
transmission  mode,  the  previously  described  axi- 
symmetric  focused  transducer  was  used  as  a  trans- 
mitter, and  a  4  mm  diameter,  10  MHz  frequency, 
PZT  cera:mic  sandwi ch-type  piezoelectric  probe  was 
used  as  a  receiver.    A  Panametrics  Model  5050R 
unit  pulse-excited  the  transmitter,  and  a  series 
unit  step  attenuator  was  used  to  attenuate  the 
ultrasonic  received  signals  prior  to  their 
entrance  into  an  amplifier  receiver.    The  breast 
preparation  was  mounted  on  a  three-motion  co- 
ordinate system  in  a  temperature  controlled  mam- 
malian Ringer's  bath  (average  temperature  23  °C) 
and  positioned  between  the  sending  and  receiving 
transducers,  with  the  anterior  aspect  of  the  tis- 
sue facing  the  sending  transducer  (fig.  1).  With 
this  arrangement,  any  area  of  the  tissue  could  be 
easily  examined  and  the  tissue  could  be  moved  in 
1  mm  steps  in  the  three-axis  coordinate  system. 

In  order  to  record  the  system's  reference  wave- 
form, transmitter  and  receiver  transducers  were 
placed  facing  each  other  and  adjusted  in  position 
to  receive  the  optimum  plane  wave  acoustic  signals. 
A  linearity  test  of  the  system  was  performed  by 
setting  the  attenuator  unit  at  different  values 
and  comparing  the  corresponding  outputs  of  the  FFT 


86 


Fig.  1.    Experimental  arrangement  of  examining 
transducer  (left),  excised  breast  (side 
view)  and  receiving  transducer. 


spectrum.    The  system  was  found  to  be  accurate  to 
±  1  dB  in  the  useable  frequency  band,  for  any 
particular  transmitting  transducer.    The  received 
ultrasonic  waveforms  were  digitized  at  a  sampling 
rate  of  10  or  20  nanoseconds  using  an  eight  bit 
Biomation  Model  8100  digitizer  and  then  transfer- 
red to  a  PDP-11/45  computer.    Generally,  four 
waveforms  were  digitized  for  each  tissue  region 
of  interest  (i.e.,  nipple,  areola,  etc.),  stored 
on  a  computer  disc,  and  later  processed  using  a 
1024  point  FFT.    In  order  to  obtain  attenuation 
and  phase-angle  versus  frequency  plots,  they  were 
deconvolved  with  the  original  system's  reference 
waveform.    The  waveforms  were  transformed  into  the 
frequency  domain  and  the  absolute  amplitude  for 
each  waveform  was  compared  with  the  amplitude  of 
the  reference  waveform  for  the  relative  attenuation 
measurement.    To  obtain  the  phase  angle  for  each 
waveform,  the  phase  was  compared  with  the  refer- 
ence waveform  and  subsequently  linearized  for  each 
frequency.    For  optimum  frequency  resolution,  ap- 
propriate sampling  intervals  were  selected  on  the 
Biomation  digitizer.    The  output  was  presented  in 
a  graphical  form  in  three  different  formats;  name- 
ly, normalized  pressure  amplitude,  attenuation  and 
phase-angle  spectra.  The  phase-angle  data  is  pre- 
sented in  a  separate  paper  [10]. 

The  tissue  was  sampled  at  various  selected  re- 
gions by  moving  the  specimen  across  the  ultra- 
sound beam  on  the  three-axis  system.    The  regions 
of  interest  chosen  for  this  study  were  specific 
tissue  paths  (i.e.,  f"om  skin  to  back  surface  of 
the  specimen)  which  included  1)  the  malignant 
tumor,  2)  the  nipple  and  3)  the  areola.    In  that 
regard,  eighteen  separate  transmission  mode  studies 
were  made  in  the  tumor  path,  nine  in  the  areola 
path  and  eight  in  the  nipple  path.    In  addition, 
the  upper  inner  aspect  of  the  breast  was  chosen  as 
representative  of  normal,  middle-aged,  mamnary 
adipose  tissue  and  test  ultrasound  transmissions 
were  made  in  that  region. 

Following  completion  of  the  FFT  studies,  the 
tissue  was  longitudinally  sliced  to  produce  whole 
breast  sections  from  5  to  8  mm  in  thickness.  Each 
of  these  cross  sections  was  x-rayed  and  the  radio- 
graphs studied  to  determine  which  specific  regions 
of  the  breast  would  be  further  examined  by  means 


of  histological  techniques.    Tissue  cubes  2  x  2.5 
cm  in  overall  dimensions  were  excised  from  the 
selected  regions  of  interest  and  prepared  for 
histological  study.    The  primary  malignant  mass 
was  included  in  one  of  these  cubes  of  tissue. 

3.  Results 

The  ongoing  histological  studies  indicate  that 
the  primary  malignant  mass  was  an  invasive  carci- 
noma of  duct  cell  origin  with  an  intermediate  de- 
gree of  differentiation.    The  mass  was  multinodu- 
lar, fairly  v/ell  circumscribed  but  not  encapsu- 
lated, with  a  dense  fibrotic  center  and  a  peri- 
pheral shell  (3  to  4  mm  thickness)  of  neoplastic 
cells.    The  fibrotic  tissue,  representing  the 
major  tumor  mass,  was  highly  collagenous  but  the 
cellular  shell  had  a  minimum  deposit  of  collagen. 

Included  in  the  illustrations  shown  in  this 
paper  are  duplications  of  the  recorded  system 
reference  waveform  (normalized  pressure  ampli- 
tude versus  frequency)  for  sound  wave  transmission 
through    Ringer's  solution,  compared  to  the  wave- 
forms recorded  for  sound  transmission  through 
breast  tissue  immersed  in  Ringer's  solution.  Such 
pressure  amplitude  graphical  displays  are  present- 
ed in  order  to  demonstrate  the  characteristics  of 
the  reference  waveform  and  the  dynamic  range  limi- 
tations of  the  system. 

Before  discussing  the  results  obtained  for  the 
tissue  path  that  included  the  malignant  tumor,  it 
is  of  interest  to  consider  the  attenuation  values 
obtained  for  the  areola  and  nipple  tissue  path  re- 
gions of  the  breast  which,  on  the  basis  of  the 
x-ray  examination,  can  be  assumed  normal.  In 
calculating  precise  attenuation  values,  on  the 
basis  of  the  recorded,  normalized  pressure  ampli- 
tude values,  the  tissue  path  length  was  taken  as 
the  distance  between  the  anterior  surface  of  the 
skin  overlying  the  nipple  or  areola  and  the  poste- 
rior surface  of  the  breast  specimen  (as  determined 
by  the  previously  obtained  ultrasound  visualiza- 
tion images  of  the  breast  specimen).  Therefore, 
it  is  emphasized  here  that  the  attenuation  values 
shown  in  figures  3  and  4  are  not  specifically  for 
areola  or  nipple,  but  are  for  the  tissue  path  that 
includes  these  structures. 

Figure  2  presents  recorded  pressure  amplitude 
waveforms  for  four  sound  transits  through  the 
nipple  at  entrance  points  separated  by  2  mm  of 
surface  tissue.    These  waveforms  show  the  maximum 
variability  in  pressure  amplitude  (differences  in 
pressure  amplitude  for  waveforms  A,  B,  C,  D)  found 
for  the  nipple  region.    The  attenuation  frequency 
spectrum  plot  for  the  nipple  path  shown  in  figure 
3  is  based  on  the  pressure  amplitude  waveforms 
shown  in  figure  2  and  the  tissue  path  length  (sur- 
face of  nipple  to  back  surface  of  the  specimen) 
traversed  by  the  sound  wave. 

Figure  4  is  the  attenuation  frequency  spectrum 
for  the  areola  region  of  the  breast.    As  in  the 
case  of  the  nipple,  this  data  is  also  based  on  dif- 
ferences in  pressure  amplitude  response  curves 
(three  sound  transits  through  the  areola  region  at 
entrance  points  separated  laterally  by  1  mm  of  sur- 
face tissue)  and  the  tissue  path  length  (areola 
skin  to  back  surface  of  the  specimen).    All  other 
sound  transmissions  in  the  areola  path  gave  re- 
sults within  the  range  of  values  shown  in  figure 
4.    The  tissue  region  of  greatest  attenuation,  as 
shown  in  figure  4,  was  located  closer  to  the  nipple 
in  comparison  to  the  other  two  sound  transit  regions. 


87 


 JL_  1 

12345678 

Frequency  Megahertz 

Fig.  2.  Computer  recorded  display  of:  (a)  ref- 
erence waveform  for  ultrasound  trans- 
mission through  Ringer's  solution  only; 
(b)  waveforms  A,  B,  C,  D  for  ultrasound 
transmission  through  four  separate  re- 
gions of  the  "nipple  tissue  path"  of  an 
excised,  formalin-fixed  breast  immersed 
in  Ringer's  solution. 

9r 


8  - 


1  - 

1         I  I  I  I  I 

1  2  3  4  5  6 

Frequency  MHz 

Fig.  3.  Attenuation  frequency  spectrum 
for  "nipple  tissue  path"  based 
on  waveforms  shown  in  figure  2. 


I        I        I        I  I  I 

1  2  3  4  5  6 

Frequency  MHz 

Fig.  4.    Attenuation  frequency  spectrum 
for  "areola  tissue  path"  based 
on  three  sound  transmissions 
through  separate  regions  of  the 
areol a . 


Frequency  Megahertz 


Fig.  5.    Computer  recorded  display  of:    (a)  ref- 
erence waveform  for  transmission  through 
Ringer's  solution  only;  (b)  three  wave- 
forms for  ultrasound  transmission  in 
separate  regions  of  the  "tumor  tissue  path" 
of  an  excised,  formalin-fixed  breast  im- 
mersed in  Ringer's  solution. 


88 


The  wavefonn  recordings  shown  in  figure  5  re- 
sulted from  three  sound  transits  in  the  tumor  tis- 
sue path,  with  each  of  the  test  points  separated 
by  1  mm  of  tumor  tissue  in  terms  of  skin  surface 
distance.    In  contrast  to  results  obtained  in  all 
other  tested  regions  of  the  breast,  the  recorded 
waveform  data  show  a  total  lack  of  recorded  pres- 
sure amplitude  response  in  the  higher  frequency 
regions.    This  result  indicates  that  there  is  at 
least  40  dB  of  amplitude  loss,  for  the  tissue  path 
from  skin  to  back  surface,  in  the  frequency  re- 
gions above  approximately  2  MHz.    At  the  lower  end 
of  the  frequency  spectrum,  pressure  amplitude  re- 
sponse was  recorded  but,  as  shown  later  in  this 
paper,  the  level  of  these  pressure  amplitudes  in- 
dicates a  greater  attenuation  for  the  tumor  tissue 
path  than  for  that  of  the  nipple  tissue  path. 

For  the  main  body  of  the  primary  malignant 
tumor,  the  overlying  tissue  is  approximately 
1.5  cm  in  depth  (from  skin  to  anterior  border 
of  overt  mass);  the  tumor  has  a  diameter  of 
1.5  cm  in  depth  and  the  tissue  below  the  tumor 
is  approximately  2  cm  in  depth.    X-ray  examina- 
tion of  the  excised  breast  indicated  the  tis- 
sues surrounding  the  tumor  were  primarily  fat- 
ty in  nature.    Consequently,  if  the  attenuation 
for  the  3.5  cm  of  tissue  surrounding  the  tumor 
is  considered  to  be  that  of  normal  breast  adi- 
pose tissue  and,  further,  if  the  attenuation 
value  for  such  tissue  is  assumed  comparable  to 
that  found  in  the  present  studies  for  tissue 
paths  (skin  to  back  of  tissue)  through  other 
fatty  regions  of  the  breast,  namely,  2.5  dB/cm 
at  2.0  MHz,  then  the  tissues  surrounding  the  tumor 
account  for  8.8  dB  of  the  attenuation.    It  should 
be  noted  that  the  2.5  dB/cm  value  for  breast  adi- 
pose tissue  is  in  general  agreement  with  that 
found  by  Calderon  et  al.  [9],  namely  2.3  dB/cm  at 
2.25  MHz  for  formalin-fixed,  excised,  normal 
breast  tissue.    Deducting  the  8.8  dB  from  the  40 
dB  maximum  dynamic  range  response  of  the  system 
and  taking  into  account  the  1.5  cm  tumor  path,  it 
is  found  that  the  attenuation  for  the  malignant 
tumor  itself  is  21  dB/cm  at  a  frequency  of  2  MHz. 
Since  this  value  is  in  general  agreement  with  that 
of  Calderon  et  a1 .  who  found  an  attenuation  of 
20  dB/cm  at  2.25  MHz  for  formalin-fixed,  excised, 
malignant  breast  tumors,  it  is  expected  that  the 
40  dB  dynamic  range  limitation,  which  represents 
a  factor  of  100  in  pressure  amplitude  and  10,000 
in  intensity,  is  close  to  an  adequate  dynamic 
range  response. 

Carrying  out  the  same  type  of  calculation,  using 
the  pressure  amplitude  data  shown  in  figure  5  for 
the  lower  frequencies,  gives  an  attenuation  value 
for  the  malignant  mass  of  8  dB/cm  at  a  frequency 
of  1.5  MHz.    This  is  calculated  on  the  basis  of 
3.5  cm  of  normal  fatty  breast  tissue,  with  an  at- 
tenuation value  of  2.0  dB/cm  at  1.5  MHz  (as  found 
for  other  normal  fatty  regions  of  this  breast 
specimen)  and  a  1.5  cm  depth  of  tumor  mass. 

Of  the  total  of  18  separate  runs  through  the 
tumor  tissue  path,  with  the  exception  of  two  sound 
transits  in  a  particular  region  of  this  path,  all 
recorded  results  indicated  that  for  frequencies 
above  2  MHz  there  apparently  was  total  attenuation 
of  the  sound  wave  (i.e.,  within  the  40  dB  record- 
ing level  capability  of  the  system).    The  excep- 
tional cases  showed  1)  a  pressure  amplitude  wave- 
form that  was  highly,  but  not  completely,  attenuat- 
ed by  the  tumor  path  and  thus  was  within  the  40  dB 
maximum  dynamic  range  of  the  system  and  2)  for  a 


sound  transit  in  an  area  just  2  mm  laterally  dis- 
tant from  that  of  number  one  above,  a  pressure 
amplitude  waveform  which  had  precisely  the  same 
frequency  spectrum,  pressure  amplitude  response 
as  that  found  for  normal  fatty  regions  of  the 
breast.    On  the  basis  of  this  result,  it  was  as- 
sumed, prior  to  the  histological  examination,  that 
the  tumor  included  a  border  region  of  less  density 
(conjectured  to  be  a  mixture  of  normal  and  neo- 
plastic tissue)  than  the  central  mass  and  an  ad- 
jacent region  of  distinct  adipose  tissue.  Based 
on  pressure  amplitude  data  and  the  known  tissue 
path,  an  attenuation  value  of  13  dB/cm  at  a  fre- 
quency of  2  MHz  was  calculated  for  this  border 
area  of  neoplastic  tissue. 

The  histological  studies,  to  date,  clearly  show 
adipose  tissue  immediately  adjacent  to  and  sur- 
rounding the  malignant  mass,  thus  accounting  for 
the  characteristics  of  the  wavefonn  which  duplicat- 
ed those  of  the  fatty  regions  of  the  breast.  The 
3  to  4  mm  thick  neoplastic  shell  located  adjacent 
to  the  adipose  tissue  and  surrounding  the  fibrotic 
center  of  the  tumor  presumably  accounted  for  the 
13  dB/cm  attenuation  value. 

4.  Discussion 

At  the  present  time,  in  clinical  studies  con- 
cerned with  the  use  of  pulse-echo  methods  for 
breast  examination,  differential  diagnosis  is  pri- 
marily based  on  the  characteristics  of  the  wall  of 
the  tumor,  the  presence  and  nature  of  the  echoes 
from  the  internal  structure  of  the  tumor  and  the 
existence  or  absence  of  a  shadowing  phenomenon  re- 
sulting from  the  attenuation  of  acoustic  energy 
by  a  malignant  mass.    In  evaluating  each  of  these 
parameters,  considerable  emphasis  and  reliance  is 
placed  on  the  so-called  "attenuation  shadow"  which, 
when  present,  is  generally  considered  to  be  indi- 
cative of  the  presence  of  a  malignant  mass.  How- 
ever, precise  knowledge  regarding  attenuation 
characteristics  of  normal,  benign  and  malignant 
breast  tissue  is  extremely  limited.    This  is  a 
serious  deficit  since  its  consequence  may  be  mis- 
diagnosis of  the  benign  or  malignant  nature  of  a 
breast  tumor. 

Considerable  care  should  be  used  in  interpret- 
ing attenuation  shadows  produced  by  pulse-echo 
examination  of  breast,  with  specific  regard  to  the 
question  of  whether  normal  or  benign  tissues  can 
produce  significant  shadowing  effects.    In  that 
regard,  Kobayashi  has  found  that  fat  necrosis  re- 
sults in  an  attenuation  shadowing  comparable  to 
that  produced  by  a  malignant  mass,  as  judged  by 
pulse-echo  visualization  [11].    Calderon  et  al . ' s 
studies  at  a  frequency  of  2.25  MHz  of  formal in- 
fixed  excised  breast  tissue  gave  values  of  20  dB/ 
cm  for  malignant  masses,  9  dB/cm  for  benign  masses 
and  approximately  2.3  dB/cm  for  normal  breast  tis- 
sue surrounding  the  tumors  [9].    Although  there 
is  a  significant  difference  between  attenuation 
values  for  benign  and  malignant  tumors  as  found  by 
these  investigators,  it  must  be  realized  that  in 
the  clinical  pulse-echo  methods,  differential  diag- 
nosis is  made  on  the  basis  of  a  comparison  between 
the  image  pattern  of  the  overt  mass  and  that  of 
the  surrounding  normal  tissue.    Therefore,  the 
Calderon  et  al .  data  indicate  that  for  pulse-echo 
techniques  in  which  the  sound  beam  makes  two  pass- 
es through  the  breast,  common  sized  benign  tumors 
(2  cm  and  above)  with  an  attenuation  of  9  dB/cm 
could  result  in  shadowing  in  relation  to  surround- 


89 


ing  normal  tissue  with  an  attenuation  value  of 
2.3  dB/cm.    Additionally,  the  first  author  of  this 
paper  recently  found,  in  the  case  of  several  clini- 
cal patients,  each  with  an  overt,  benign  mass 
within  the  breast  (identified  by  x-ray  examination 
and  confirmed  by  surgical  biopsy),  that  the  echo- 
grams obtained  using  standard  pulse-echo  methods 
at  frequencies  of  4.4  MHz  to  5.0  MHz  displayed 
distinct  shadows  in  the  region  of  the  mass.  The 
significance  of  this  result,  insofar  as  the  present 
paper  is  concerned,  is  the  recognition  of  the  need 
for  quantitative  attenuation  data  on  multiple 
types  of  benign  and  malignant  breast  tumors,  as 
opposed  to  simple  visual  inspection  of  shadow 
phenomena  of  such  masses  and  consider.  -^ion  of 
whether  signal  processing  techniques  may  orovide 
some  of  the  needed  data  [12]. 

The  FFT  studies  discussed  in  this  paper,  for 
the  tissue  path  which  included  the  malignant  tumor, 
showed  that  the  attenuation  for  the  full  range  of 
frequencies  studied  (1.1  to  4.4  MHz)  was  greater 
for  this  tissue  path  than  for  that  of  any  other 
area  of  the  breast.    Further,  these  findings  were 
consistent  with  those  found  in  the  visualization 
echograms,  that  is  to  say,  even  at  a  frequency  of 
1.1  MHz,  this  specific  tumor  could  be  recognized 
by  "pale  shadowing,"  while  at  the  higher  frequen- 
cies the  attenuation  shadow  was  more  opaque  and 
formed  a  distinctive  border  in  relation  to  ad- 
jacent tissue. 

From  the  viewpoint  of  possible  in  vivo  breast 
examinations  by  FFT  methods,  it  is  of  interest 
that  in  the  present  study  the  dual  structural  com- 
ponents of  the  tumor  (as  shown  by  the  two  attenua- 
tion values)  and  the  adipose  tissue  surrounding 
the  tumor  were  recognized  prior  to  any  histologi- 
cal investigations.    Further,  the  finding  that 
the  tc'  ^1  malignant  mass  was  attenuating  but  the 
greater  attenuation  was  in  the  region  of  fibrosis 
agrees  with  earlier  studies  [8]  relating  attenua- 
tion (as  judged  by  nonquantitati ve  pulse-echo 
methods)  of  breast  tumors  with  specific  tumor  tis- 
sues (as  revealed  by  detailed  histological 
studies).    The  highly  collagenous  nature  of  the 
primary  malignant  mass  combined  with  its  high  at- 
tenuation values  and  the  minimum  collagen  content 
of  the  surrounding  cellular  shell  is  of  interest 
in  regard  to  the  relative  importance  of  elastic 
properties  versus  physical  structural  make-up  for 
sound  reflection  and  transmission  through  overt 
tissue  masses  [13].    The  ongoing  histological 
studies  and  the  preliminary  phase-angle  data  may 
provide  some  information  on  this  aspect  [10]. 

There  are  many  factors,  both  tissue  and  instru- 
mentation related,  associated  with  the  attenuation 
in  the  nipple  path  region  as  determined  by  the  FFT 
ultrasound  transmission  studies  discussed  in  this 
paper.    Included  in  the  tissue  aspects  are  the  non- 
uniform structure  of  the  skin  surface  and  the  vari- 
able tissue  components  within  the  nipple  and  in 
the  regions  deep  to  the  nipple.    Although  the 
above  factors  may  be  significant,  information 
which  is  relevant  to  present  pulse-echo  clinical 
techniques  can  be  derived  from  the  preliminary 
FFT  data  without  complete  analysis  of  such  param- 
eters.   In  that  regard,  for  example,  the  results 
shown  in  figure  3  are  important  in  relation  to 
detection  of  an  overt  breast  mass  located  in  the 
region  directly  posterior  to  the  nipple.    If  pulse- 
echo  techniques  are  applied  to  scan  directly  over 
the  nipple  region,  there  may  be  a  lack  of  success 
in  detecting  such  masses  (particularly  at  frequen- 


cies of  the  order  of  5  MHz)  because  of  their  loca- 
tion in  the  attenuating  path  of  the  nipple.  For 
the  purposes  oftumor  detection,  therefore,  the 
tissues  deep  to  the  nipple  should  be  viewed  by 
scanning  from  the  side  regions  of  the  breast  and 
by  using  lower  frequencies.    These  two  approaches 
to  visualizing  structures  deep  to  the  nipple  were 
successfully  applied  by  the  present  authors  in  the 
case  of  the  excised  breast  discussed  in  this  paper, 
and  breasts  studied  in  vivo. 

To  accomplish  the  overall  aim  of  the  studies 
carried  out  in  this  preliminary  investigation, 
that  is,  to  correlate  x-ray,  visualization,  FFT 
analysis  and  histological  findings,  requires  ex- 
tensive and  time-consuming  investigations.  How- 
ever, if  some  of  the  significant  parameters  that 
are  relevant  to  differential  diagnosis  of  breast 
tumors  by  ultrasonic  techniques  are  to  be  deter- 
mined by  methods  other  than  the  necessai^'iy  slow 
process  of  interpreting  ultrasound  clinical  data, 
such  detailed  investigations  are  necessary.  A 
significant  result  of  the  investigation  reported 
in  this  paper  is  the  determination  that  this  com- 
puter based,  signal  processing  technique  is  feas- 
ible and  can  yield  data  on  malignant  and  normal 
regions  of  breast  tissue.    It  also  appears  evident 
from  these  experiments  that  similar  investigations 
could  be  carried  out  on  freshly  excised  whole 
breast  tissue.    If  such  investigations  are  suc- 
cessful, then  this  technique  could  be  considered 
for  application  to  breast  patient  examination. 
Success  in  that  regard  could  lead  directly  to  de- 
velopment of  an  ultrasound  breast  examination 
technique  which  might  eventually  limit  the  need 
for  surgical  biopsy  to  a  small  number  of  excep- 
tional cases. 

Acknowledgments 

Grateful  acknowledgment  is  made  to  the  assist- 
ance of  George  W.  Ga    ler,  ultrasound  technologist, 
in  all  aspects  of  the  above  studies. 

This  research  was  supported  by  the  Grace  M. 
and  Ralph  W.  Showalter  Residua»"y  Trust,  and  by 
the  Indianapolis  Center  for  Advanced  Research,  Inc. 

References 

[1]    Kobayashi,  T.,  Review:    Ultrasonic  Diagnosis 
of  Breast  Cancer,  in  Ultrasound  in  Medicine 
and  Biology,  Vol.  1  (1975),  pp.  383-391. 

[2]    Gallager,  H.  S.  and  Martin,  J.  E.,  The  Patholo- 
gy of  Early  Breast  Cancer,  in  ' ~east  Cancer: 
Early  and  Late,  The  Univers:       f  Texas, 
M.  D.  Anderson  Hospital,  pp.  3/-50  (Year  Book 
Medical  Publishers,  Inc.,  Chicago,  1968). 

[3]    Gallager,  H.  S.  and  Martin,  J.  M.,  The  study 
of  mammary  carcinoma  by  mammography  and  whole 
organ  sectioning.  Cancer  23  (4),  855-873 
(1969). 

[4]    Fry,  E.  Kelly,  Franklin,  T.  D.,  Jr.,  and 

Gallager,  H.  S.,  Ultrasound  Visualization  of 
Excised  Breast  Tissue:    An  Experimental  Ap- 
proach to  the  Problem  of  Precise  Identifica- 
tion of  Structure  from  Echogram  Data,  Acous- 
tical Society  of  America  meeting,  Washington, 
D.C.,  April  (1971). 

[5]    Fry,  E.  Kelly,  Gallager,  H.  S.,  and  Franklin, 


90 


T.  D. ,  Jr. ,  In  Vivo  and  In  Vitro  Studies  of 
Application  o7  Ultrasonir~Visualization  Tech- 
niques for  Detection  of  Breast  Cancer,  in 
Proc,  IEEE  Ultrasonics  Symposium,  Miami, 
Florida,  December  (1971).  ~ 

[6]    Martin,  J.  E.  and  Gallager,  H.  S.,  Reflec- 
tions on  Benign  Disease:    A  Radiographic- 
Histologic  Correlation,  in  Early  Breast 
Cancer  Detection  and  Treatment,  H.  S.  Gal- 
lager, ed. ,  pp.  177-181  (John  Wiley  and  Sons, 
New  York,  1976). 

[7]    Fry,  E.  Kelly,  The  Use  of  Ultrasound  Methods 
to  Detect  Changes  in  Breast  Tissue  Which  Pre- 
cede the  Formation  of  a  Malignant  Tumor,  in 
Acoustical  Holography,  L.  W.  Kessler,  ed., 
pp.  1-20,  Vol.  7  (Plenum  Publishing  Corp., 
New  York,  1977). 

[8]    Fry,  E.  Kelly  and  Gallager,  H.  S.,  A  Research 
Approach  to  Visualization  of  Breast  Tumors  by 
Ultrasound  Methods,  in  Ultrasound:    Its  Ap- 
plication in  Medicine  and  Biology,  F.  J.  Fry, 
ed.  (Elsevier  Scientific  Publishing  Co. , 
Amsterdam,  1978). 

[9]    Calderon,  C,  Vilkomerson,  D.,  Mezrich,  R., 
Etzold,  K.  F.,  Kingsley,  B.,  and  Haskin,  M. , 
Differences  in  the  attenuation  of  ultrasound 


by  normal,  benign  and  malignant  breast  tis- 
sue, J.  Clin.  Ultrasound  4  (4),  249-254 
(19761: 

[10]    Fry,  E.  Kelly,  Sanghvi ,  N.  T.,  Fry,  F.  J., 
Gardner,  G.  W.,  and  Gallager,  H.  S.,  Deter- 
mination of  Alterations  of  Phase  Angle  of 
Ultrasound  Transmitted  Through  a  Malignant 
Breast  Tumor:    A  Preliminary  Investigation, 
in  Ultrasound  in  Medicine,  D.  White,  ed.. 
Vol.  4  (Plenum  Publ ishing  Corp.,  New  York, 
1978). 

[11]    Kobayashi,  T.,  Takatani,  0.,  Hattori,  N., 
and  Kimura,  K.,  Study  of  sensitivity-graded 
ultrasonotomography  of  breast  tumor  (pre- 
liminary report),  Med.  Ultrasonics  10  (1), 
38-40  (1972). 

[12]    Lizzi,  F.,  Katz,  L.,  St.  Louis,  L.,  and 

Jackson-Coleman,  D.,  Applications  of  spectral 
analysis  in  medical  ultrasonography.  Ultra- 
sonics 14  (2),  77-80,  March  (1976). 

[13]    Fields,  S.  and  Dunn,  F.,  Correlation  of 

echographic  visual izabil ity  of  tissue  with 
biological  composition  and  physiological 
state,  J.  Acoust.  Soc.  Amer.  54  (3),  809- 
812  (1973T: 


91 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


CORRELATION  OF  ULTRASONIC  ATTENUATION  WITH  CONNECTIVE  TISSUE 
CONTENT  IN  BREAST  CANCERS 

Toshiji  Kobayashi 

Division  of  Clinical  Electrophysiology 
Department  of  Internal  Medicine 
National  Cancer  Center  Hospital 
5-1-1,  Tsukiji,  Chuo-ku,  Tokyo-104,  Japan 


In  recent  years,  ultrasonic  techniques  have  been  applied  to  the  diagnosis  of  breast 
cancer.    Various  characteristics  suggestive  of  tumor  pathology  have  been  determined. 
In  this  paper,  some  of  these  signs,  such  as  the  acoustic  middle  shadow  and  the  com- 
plete disappearance  of  the  distal  limit  of  the  tumor  mass  echo,  are  correlated  with 
the  connective  tissue  content  of  various  breast  cancers.    Because  of  the  high  acoustic 
impedance  of  connective  tissue  relative  to  the  other  tissues,  tumors  rich  in  connec- 
tive tissue  are  believed  more  strongly  to  attenuate  the  ultrasonic  waves  and  thus  to 
cause  shadows  in  the  tissues  beyond  them. 

Key  words:    Attenuation;  breast  cancer;  cancer;  connective  tissue;  differential 
diagnosis;  medullary  carcinoma;  papillary  carcinoma;  scirrhous 
carcinoma;  shadowing;  ultrasound. 


1.  Introduction 

Because  of  its  noninvasive  nature,  diagnostic 
ultrasound  is  a  good  clinical  tool  for  the  fine 
visualization  of  soft  tissue  pathologies.  This 
technique  has  been  applied  in  the  past  several 
years  to  the  differential  diagnosis  of  breast 
tumors  with  high  diagnostic  accuracies.  Echo- 
graphic  criteria  for  differentiating  malignant 
and  benign  lesions  have  been  reported  by  several 
investigators  [1-8]^. 

Echographic  patterns  of  shadowing  beyond  tumors 
(retromammary  shadowing)  provide  reliable  and  ac- 
curate diagnostic  information  for  the  diagnosis  of 
breast  cancer.    In  this  paper,  various  malignant 
signs,  such  as  the  acoustic  middle  shadow  sign  and 
the  complete  disappearance  of  the  distal  limit  of 
the  tumor  mass  echo,  are  shown  to  be  correlated 
with  the  connective  tissue  content  within  the 
breast  cancer.    Because  of  the  high  acoustic  im- 
pedance of  connective  tissue  relative  to  the  other 
tissue,  tumors  rich  in  connective  tissue  are  be- 
lieved more  strongly  to  attenuate  the  ultrasonic 
waves  and  thus  to  cause  shadows  in  the  tissues  be- 
yond them. 

2.    Method  and  Results 

During  the  past  five  years,  1617  cases  of  pal- 
pable breast  tumors  were  examined  echographi cal ly 
and  all  cases  with  breast  cancer  were  verified  by 
mastectomy.    This  material  forms  the  basis  of  this 
retrospective  study.    The  method  of  examination  is 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


shown  in  figure  1.    The  criteria  for  the  differen- 
tial diagnosis  of  breast  tumors  can  be  divided  in- 
to three  major  categories  comprising  (1)  the  bound- 
ary echoes  and  shape,  (2)  internal  echoes,  and 
(3)  shadowing,  as  illustrated  in  figure  2. 


Fig.  1.    Method  of  scanning  the  breast.    The  5  MHz 
probe  is  moved  automatically  along  the  arc 
of  a  circle,  and  ultrasonic  coupling  to  the 
patient  is  through  the  water  in  the  flexible 
plastic  bag. 


93 


BOUNDARY  ECHO  &  SHAPE 


usual ly  regular  and 
smooth,  round,  oval 
or  hetnioval 


INTERNAL  ECHO 


RETROMAMMARY  SHADOWING 


uniform- si  zed, 
homogeneous  or 
echo-free  (anechoicl 


irregular  and  jagged 
bizarre,  crab-like  or 
polymorphous 


non-uniform-sized , 
heterogeneous  or 
polymerous 


tadpole-tail  sign 
lateral  shadow  sign 
accentuation  of  posterior  echo 


acoustic  middle  shadow 
(posterior  shadowing) 
attenuation  of  posterior  echo 


Fig.  2.    Diagram  illustrating  echogram  appearances  for  differential  diagnosis 
of  benign  and  malignant  tumors. 


The  first  two  categories,  that  is,  the  boundary 
echoes  and  shape,  and  internal  echoes,  result  from 
echo  patterns  mainly  confined  to  the  tumor  mass  it- 
self or  its  vicinity.    The  third  category,  shadow- 
ing, is  a  result  of  various  attenuation  mechanisms 
such  as  multiple  reflections,  changes  in  beam  velo- 
city, and  scattering  due  to  the  impedance  discon- 
tinuities within  the  tumor. 

The  boundary  echoes  provide  important  diagnostic 
information,  producing  outlines  which  are  usually 
irregular  in  the  case  of  breast  cancers,  and  regu- 
lar in  the  case  of  benign  lesions,  such  as  cysts 
and  fibroadenomas.    A  cancerous  tumor  is  uneven 
with  a  jagged  edge,  sometimes  triangular,  rectangu- 
lar or  irregular  in  shape,  whereas  a  benign  tumor 
is  usually  round,  oval  or  hemioval.    Moreover,  the 
internal  echo  pattern  in  a  malignant  tumor  is 
usually  irregular  both  in  size  and  distribution,  in 
contrast  with  the  either  homogeneous  or  anechoic 
internal  echo  pattern  of  a  benign  lesion. 

Other  differential  diagnostic  criteria  result 
from  echo  patterns  due  to  multiple  reflections  from 
the  tumor  mass,  modifying  the  shadowing.    It  is 
this  phenomenon  which  is  the  principal  subject  of 
this  paper.    The  tadpole-tail  sign  arises  from  the 
low  energy  loss  in  the  beam  as  it  passes  normally 
through  the  distal  wall  of  a  cyst  leaving  suffi- 
cient energy  to  cause  multiple  reflections  or  "ring- 
ing" between  the  distal  cyst  wall  and  the  chest 
wall.    The  difference  in  the  velocities  in  tissue 
and  cyst  may  play  an  important  role  in  causing  this 
pattern.    This  sign  is  usually  not  seen  in  the 
majority  of  malignant  lesions  except  in  some  cases 
of  medullary  carcinoma.    The  lateral  shadow  sign  is 
believed  to  be  formed  by  the  almost  total  reflec- 
tion of  the  ultrasonic  beam  at  the  lateral  wall  of 
cystic  lesions  as,  for  example,  with  benign  tumors 
[6];  an  explanation  of  the  formation  of  this  sign 


has  recently  been  published  [9]. 

A  characteristic  of  malignant  lesions  designated 
as  "acoustic  middle  shadow  sign"  is  sometimes  ob- 
served.   This  phenomenon  is  believed  to  be  caused 
by  the  high  attenuation  of  the  ultrasonic  energy  by 
malignant  tumor  tissue,  due  to  large  acoustic  im- 
pedance discontinuities  within  such  tissue. 

The  reliability  of  these  various  signs  for  the 
differential  diagnosis  of  various  histological  types 
of  breast  cancer  is  seen  in  table  1.  Diagnostic 
accuracy  is  greatest  in  scirrhous  carcinoma  and 
lowest  in  medullary  carcinoma. 

Echographic  characteristics  suggestive  of  malig- 
nant lesions,  such  as  the  acoustic  middle  shadow 
sign  and  the  complete  disappearance  of  the  distal 
limit  of  the  tumor  mass  echo,  have  been  correlated 
with  the  connective  tissue  content  of  tumor  masses 
for  various  histological  types  of  breast  cancer. 
The  grading  of  the  connective  tissue  content  was  di- 
vided into  three  categories  based  on  the  observation 
of  many  microscopic  field  specimens.    These  cate- 
gories were  expressed  as  follows:    1)  the  connective 
tissue  over  75  percent  (rich),  2)  25  to  75  percent 
(moderate)  and  3)  less  than  25  percent  (poor).  The 
specimens  were  stained  by  hemtoxylin  and  eosin  and 
are  shown  magnified  by  approximately  100. 

The  results  of  the  analysis  are  shown  in  tables 
1,  2,  3  and  4.    The  acoustic  middle  shadow  sign  as 
well  as  the  complete  disappearance  of  the  distal 
limit  of  the  tumor  mass  echo,  were  most  frequently 
seen  in  the  cases  with  abundant  connective  tissue, 
whereas  the  tadpole-tail  sign  and  the  lateral 
shadow  sign  were  often  seen  in  cases  with  little 
connective  tissue  content,  mainly  in  medullary 
carcinoma.    These  analyses  were  carried  out  on  12 
cases  of  various  breast  cancers  (4  scirrhous  car- 
cinomas, 5  papillary  carcinomas  and  3  medullary 
carcinomas). 


94 


Table  1.    Reliability  of  various  criteria  for  the  differential  diagnosis 
of  breast  cancer  (53  cases). 


4->  O 


ro  -r- 

s-  -c 

ra 

01  S- 
Q-  O 
Q-  E 
=C  3 


Type  of  carcinoma 

Medullary 

Papi 1 1 ary 

Sci  rrhous 

Diagnostic  accuracy  rate 

20/24 

(83°/,) 

13/15 

(87%) 

14/14 

(100%) 

nant 

Complete  disappearance  of 
distal  limit  of  tumor  echo 

17/24 

(70%) 

12/15 

(80%) 

11/14 

f  -t  no/  \ 

(78%) 

Irregular  boundary  echo 

23/24 

(95%) 

14/15 

(93%) 

11/14 

(78%) 

ID 

s: 

Acoustic  middle  shadow  sign 

18/24 

(75%) 

11/15 

(73%) 

14/14 

(100%) 

c 

Bilateral  disappearance  of 
distal  limit  of  tumor  echo 

4/24 

(17%) 

1/15 

(0.6%) 

0/14 

(0%) 

(U 

Tadpole-tail  sign 

4/24 

(17%) 

2/15 

(1.3%) 

0/14 

(0%) 

CO 

Lateral  shadow  sign 

4/24 

(1.3%) 

2/15 

(1.3%) 

0/14 

(0%) 

Table  2.    Connective  tissue  content  of  12  cases  of  carcinoma.    Rich  =  over  75  percent; 

moderate  =  25  to  75  percent;  poor  =  less  than  25  percent.    High  consistency  is 
indicated  by  (+++),  intermediate  consistency  by  (++),  and  low  consistency  by  (+). 


Connective  tissue  content 


Nonconnecti ve  tissue  content 


1) 

Scirrhous 

carcinoma 

(Tl) 

(+++) 

2) 

Sci  rrhous 

carcinoma 

(Tl) 

(+++) 

3) 

Scirrhous 

carcinoma 

(T2) 

(+++) 

4) 

Scirrhous 

carcinoma 

(T2) 

5) 

Papi 1 1 ary 

carcinoma 

(Tl) 

6) 

Papillary 

carcinoma 

(Tl) 

7) 

Papi 1 1 ary 

carcinoma 

(T2) 

8) 

Papi 1 lary 

carcinoma 

(T2) 

(+++) 

9) 

Papi 1 1 ary 

carcinoma 

(T2) 

(+++) 

10) 

Medul lary 

carcinoma 

(Tl) 

(+) 

11) 

Medul lary 

carcinoma 

(T2) 

(+) 

12) 

Medullary 

carcinoma 

(T2) 

(+) 

rich 


moderate 


(++) 

(++) 
(++) 
(++) 


poor 


(+++) 
(+++) 
(+++) 


rich 


moderate 


(++) 

(++) 
(++) 
(++) 


poor 

(+) 
(+) 
(+) 


(+) 
(+) 


Table  3.    Correlation  between  appearance  of  various  echo  signs  and  connective  tissue  contents  in 
12  cases  of  carcinoma,  classified  according  to  echo  signs.    Signs  indicated  by  circled 
letters  in  the  table  have  the  higher  reliabilities.    A  =  acoustic  middle  shadow  sign; 
C  =  complete  disappearance  of  distal  limit  of  tumor  mass  echo;  N=  no  change  of  shadow 
or  intermediate  pattern;  T  =  tadpole-tail  sign;  L  =  lateral  shadow  sign. 

Connective  tissue  content 


rich  moderate  poor 


1) 

Scirrhous 

care i  noma 

(Tl) 

(A) 

N 

T 

L 

A 

C 

N 

T 

L 

A 

C 

N 

T 

L 

2) 

Scirrhous 

carcinoma 

(Tl) 

@ 

N 

T 

L 

A 

C 

N 

T 

L 

A 

C 

N 

T 

L 

3) 

Scirrhous 

carcinoma 

(T2) 

C 

N 

T 

L 

A 

(0 

N 

T 

L 

A 

C 

N 

T 

L 

4) 

Sci  rrhous 

carcinoma 

(T2) 

A 

c 

N 

T 

L 

A 

c 

N 

'X) 

CD 

A 

C 

N 

T 

L 

5) 

Papillary 

carcinoma 

(Tl) 

A 

c 

N 

T 

L 

A 

c 

N 

(T) 

A 

C 

N 

T 

L 

6) 

Papi 1 lary 

carcinoma 

(Tl) 

A 

c 

N 

T 

L 

A 

c 

N 

(T) 

® 

A 

C 

N 

T 

L 

7) 

Papi 1 lary 

carcinoma 

(T2) 

A 

c 

N 

T 

L 

A 

c 

IN) 

T 

L 

A 

C 

N 

T 

L 

8) 

Papil lary 

carcinoma 

(T2) 

(A) 

© 

N 

T 

L 

A 

c 

N 

T 

L 

A 

C 

N 

T 

L 

9) 

Papil lary 

carcinoma 

(T2) 

CA) 

© 

N 

T 

L 

A 

c 

N 

T 

L 

A 

C 

N 

T 

L 

10) 

Medullary 

carcinoma 

(Tl) 

A 

c 

N 

T 

L 

A 

c 

N 

T 

L 

A 

C 

N 

8 

(T) 

11) 

Medul 1 ary 

carcinoma 

(T2) 

A 

c 

N 

T 

L 

A 

c 

N 

T 

L 

A 

C 

N 

0 

12) 

Medullary 

carcinoma 

(T2) 

A 

c 

N 

T 

L 

A 

c 

(N) 

T 

L 

A 

C 

N 

T 

L 

95 


Table  4.    Correlation  between  appearance  of  various  echo  signs  and  connective  tissue 
contents  in  12  cases  of  carcinoma,  classified  according  to  carcinoma  type. 
(S)  =  scirrhous  carcinoma;  (P)  =  papillary  carcinoma;  (M)  =  medullary 
carcinoma. 


Connective  tissue  content  Nonconnecti ve  tissue  content 


rich 

moderate 

poor 

rich 

moderate 

poor 

Acoustic  middle  shadow  sign 

(S) 
(P) 

(S)  (S) 
(P) 

(S) 
(P) 

(S)  (S) 
(P) 

Complete  disappearance  of  dis- 
tal limit  of  tumor  mass  echo 

(S) 

(P) 

(S)  (P) 

(S) 

(S) 

(S) 
(P) 

(S)  (P) 

Tadpole-tail  sign 

(P)  (P) 

(M)  (M) 

(M)  (M) 

(P)  (P) 

Lateral  shadow  sign 

(P)  (P) 

(M)  (M) 

(M)  (M) 

(P)  (P) 

No  shadow  or  intermediate 
pattern 

(P)  (M) 

(P)  (M) 

3.    Demonstration  of  Typical  Echograms 
and  Tissue  Characteristics 

Tables  3  and  4  show  the  correlations  between 
the  amount  of  connective  tissue  and  the  various 
echo  patterns  described  in  the  following  para- 
graphs . 

A.    Scirrhous  Carcinoma  (Tl  and  T2) 

A  series  of  echograms  recorded  by  the  sensitivity- 
graded  method  [6]  show  typical  patterns  of  shadow- 
ing (acoustic  middle  shadow  sign)  suggestive  of  T2 
malignancy  (fig.  3). 

The  distal  limit  of  the  tumor  mass  echo  gradual- 
ly fades  as  the  attenuation  increases,  as  may  be 


Fig.  3.    Echograms  of  breast  with  T2  scirrhous 
carcinoma  (acoustic  middle  shadow  sign) 
recorded  at  a  range  of  sensitivities,  to- 
gether with  corresponding  histological 
section  magnified  approximately  100  times. 


seen  on  the  echograms  taken  at  -15  to  -20  dB  at- 
tenuation, and  finally  completely  disappears.  The 
microscopic  picture  shows  the  tissue  type  content 
to  be  predominantly  connective  in  cellular  struc- 
ture.   The  echogram  in  figure  4  is  a  typical  pat- 
tern of  early  scirrhous  carcinoma  (Tl),  showing  the 
acoustic  middle  shadow  underneath  the  tumor  mass 
echo.    Histologically,  the  connective  tissue  com- 
ponent is  predominant. 


Fig.  4.    Echogram  of  breast  with  Tl  scirrhous 

carcinoma  (acoustic  middle  shadow  sign), 
together  with  corresponding  histological 
section  magnified  approximately  100  times. 


96 


Fig.  5.    Echogram  of  breast  suggesting  Tl  scirrhous 
carcinoma  (acoustic  middle  shadow  sign), 
together  with  confirmatory  corresponding 
histological  section  magnified  approxi- 
mately 100  times. 

The  echogram  in  figure  5  shows  another  typical 
acoustic  middle  shadow  pattern  suggestive  of  Tl 
malignancy.    The  microscopic  picture  also  shows 
rich  connective  tissue. 

These  results  suggest  that  rich  connective  tissue 
content  may  play  an  important  role  in  producing  the 
acoustic  middle  shadow  sign  and  in  the  complete 
disappearance  of  the  distal  limit  of  the  tumor  mass 
echo  in  clinical  echograms  of  scirrhous  carcinoma. 

B.  Papillary  Carcinoma  (Tl) 

The  echogram  of  early  papillary  carcinoma  shows 
a  typical  attenuation  of  the  shadowing,  so  that  the 
acoustic  middle  shadow  sign  is  not  clear.  More- 
over, the  typical  benign  signs  such  as  tadpole-tail 
sign  and  the  lateral  shadow  sign  do  not  appear 
either  (fig.  6. ) . 

The  histological  picture  shows  connective  tissue 
and  nonconnecti ve  tissue  approximately  evenly  dis- 
tributed.   Therefore,  this  pattern  of  papillary 
carcinoma  may  lie  in  between  those  of  scirrhous 
carcinoma  and  medullary  carcinoma  (see  C  below) 
with  respect  to  the  connective  tissue  content. 
This  kind  of  evenly-distributed  connective  and  non- 
connective  tissue  produces  an  equivocal  echographic 
pattern  which  makes  it  difficult  correctly  to  diag- 
nose early  papillary  carcinomas. 

C.  Medullary  Carcinoma  (T2) 

This  echogram  shows  a  rather  homogeneous  in- 


Fig.  6.    Echogram  of  breast  with  papillary  carci- 
noma (note  absence  of  acoustic  middle 
shadow  and  tadpole-tail  signs),  together 
with  corresponding  histological  section 
magnified  approximately  100  times. 

ternal  echo,  mimicking  a  benign  cyst,  and  an  atypi- 
cal tadpole-tail  shorter  than  that  of  tadpole-tail 
usually  seen  in  the  case  of  benign  tumors  (fig.  7). 
Consequently  correct  diagnosis  is  rather  difficult. 
The  histological  picture  shows  rich  non-connective 
tissue  content,  and  especially  abundant  and  uni- 
formly distributed  tumor  cells.    This  kind  of 
homogeneous  and  even  distribution  of  tumor  cells 
should  produce  little  sonic  reflection  from  within 
the  tumor  mass,  thus  mimicking  the  pattern  of 
benign  cysts. 

4.  Discussion 

In  the  past  several  years,  researchers  have 
hypothesized  that  the  ultrasonic  characteristics 
of  malignant  breast  tumors  could  be  attributed  to 
the  increased  attenuation  of  ultrasonic  energy  by 
the  cancerous  tissue  with  greater  internal  acoustic 
impedance  discontinuities,  and  hence  increased 
scattering  losses,  as  compared  with  those  with 
normal  breast  tissue  and  benign  breast  tissues. 

In  direct  measurements  of  the  attenuation  of 
ultrasound  within  18  samples  of  various  normal, 
benign  and  malignant  breast  tissues,  significant 
differences  have  been  found  [10].    At  2.25  MHz, 
malignant  tissue  had  the  highest  attenuation  (1.2 
dB  per  wavelength),  whilst  benign  tissues  had  a 
median  value  of  0.6  dB  per  wavelength.    These  re- 
sults support  the  hypothesis. 

From  a  histological  standpoint,  the  percentage 
of  connective  tissue  (consisting  of  fibroblasts 


97 


Fig.  7.    Echogram  of  breast  with  T2  medullary 
carcinoma  (the  tadpole-tail  sign  is 
shorter  than  that  generally  seen  with 
benign  tumors),  together  with  correspond- 
ing histological  section  magnified  ap- 
proximately 100  times. 

and  fibrous  tissue)  has  been  suggested  to  be  one 
of  the  principal  factors  responsible  for  ultrasonic 
attenuation  in  neoplastic  tissue  [11]. 

In  an  analysis  [12]  of  53  cases  of  breast  cancer 
for  the  incidence  of  the  appearance  of  various  echo- 
graphic  characteristics  such  as  the  acoustic  middle 
shadow  sign,  the  complete  disappearance  of  the 
distal  limit  of  the  tumor  mass  echo  was  suggestive 
of  malignancy,  and  the  tadpole-tail  sign  and  the 
lateral  shadow  sign  was  suggestive  of  a  benign  con- 
dition.   Diagnostic  accuracy  rates  were  100  per- 
cent, 87  percent  and  83  percent  in  scirrhous  carci- 
noma, papillary  carcinoma  and  medullary  carcinoma 
respectively,  as  may  be  seen  in  table  1.    It  is  in- 
teresting to  note  that  the  tadpole-tail  sign  and 
the  lateral  shadow  sign  appeared  most  rarely,  and 
the  disappearance  of  the  distal  limit  of  the  tumor 
echo  was  most  frequent,  in  medullary  carcinomas. 
Generally  these  signs  suggest  benign  lesions,  and 
this  may  account  for  the  fact  that  the  lowest  diag- 
nostic accuracy  is  obtained  with  medullary  carci- 
nomas . 

The  connective  tissue  content  seems  to  be  gener- 
ally rich  in  scirrhous  carcinoma  and  poor  in  medul- 
lary carcinoma.    The  tumor  cellular  component 
usually  predominates  Within  the  nonconnecti ve  tis- 
sue of  medullary  carcinoma  and  this  tissue  struc- 
ture may  be  responsible  for  the  bioacoustical  homo- 
geneity observed  within  the  tumor  mass.    This  homo- 
geneity in  medullary  carcinoma  may  produce  less  at- 


tenuation of  ultrasonic  energy,  resulting  in  shadow- 
ing suggestive  of  benign  tumors,  such  as  the 
tadpole-tail  sign  and  the  lateral  shadow  sign.  In 
the  present  investigation,  these  echographic  charac- 
teristics appear  regardless  of  the  tumor  size  and 
the  location  of  the  tumor  mass  within  the  breast. 

Medullary  carcinoma  has  lower  attenuation  than 
other  types  of  carcinoma  [10];  its  measured  value 
at  2.25  MHz  falls  in  a  range  comparable  to  that  of 
benign  tumor  tissue,  whereas  the  attenuation  is 
highest  in  scirrhous  carcinomas  followed  by  papil- 
lary carcinomas.    It  is  interesting  to  note  that 
the  discrimination  on  the  basis  of  attenuation  was 
very  clear  between  normal  breast  tissue,  benign 
tumors,  and  malignant  tumors,  except  for  the  case 
of  medullary  carcinoma.    This  finding  is  in  agree- 
ment with  the  hypothesized  echographic  origin  of 
the  medullary  carcinomas  even  at  the  frequency  of 
5  MHz  used  in  the  present  analysis. 

As  to  the  other  factors  contributing  to  the 
echographic  characteristics  of  tumors,  especially 
irregular  and  nonsmooth  boundary  echoes  of  various 
carcinomas,  the  incident  rate  of  its  irregularity 
is  95  percent  for  medullary  carcinoma,  93  percent 
for  papillary  carcinoma  and  78  percent  for  scirrhous 
carcinoma.    This  kind  of  irregularity  may  be  close- 
ly related  with  infiltrative  pattern  of  cancerous 
cells,  and  the  reflection  coefficient  at  the  tumor 
interface  as  well  as  other  factors. 

It  should  be  emphasized  that  the  correlation  of 
clinical  echographic  signs  with  the  histological 
characteristics  of  the  tumors  does  not  imply  that 
in  any  given  case,  a  particular  sign  is  indicative 
of  any  histological  type.    Because  of  inadequacies 
of  the  clinical  equipment  used  in  this  study,  these 
correlations  should  only  be  considered  qualitative. 
The  principal  conclusion  is  that  the  amount  of  con- 
nective tissue  within  the  tumor  mass  may  play  an 
important  role  in  the  formation  of  the  acoustic 
middle  shadow  sign  and  the  complete  disappearance 
of  the  distal  limit  of  the  tumor  mass  echo,  and 
that  it  is  probably  due  to  the  increased  ultrasonic 
attenuation  caused  by  the  high  acoustic  impedance 
of  the  connective  tissue. 

References 

[1]    Baum,  G.,  Ultrasonic  Examination  of  the 
Breast,  in  Fundamentals  of  Medical  Ultra- 
sonography, G.  Baum,  ed.,  pp.  380-402 
(Putnam' s  Sons,  New  iork,  1975). 

[2]    Cole-Beuglet,  C.  and  Beique,  R.  A.,  Con- 
tinuous ultrasound  B-scanning  of  palpable 
breast  masses.  Radiology  117,  123-128  (1975). 

[3]    Fujii,  T.  ,  Izuo,  M.,  Kishi,  S.,  Yokomori ,  T., 
and  Fujimori,  M.,  The  results  and  the  evalua- 
tion of  ultrasonic  diagnosis  of  breast  dis- 
ease, J.  Jap.  Soc.  Cancer  Therap.  8^,  253- 


[4]  Hirose,  M.  and  Furuki,  R. ,  Ultrasonic  diag- 
nosis of  breast  disease,  Proc.  17th  Meeting 
Jap.  Soc.  Ultrasonics  Med.  17,  37-38  (1971) 

[5]    Jellins,  J.,  Kossoff,  G.,  Reeve,  T.  S.,  and 
Barraclough,  B.  H.,  Ultrasonic  grey  scale 
visualization  of  breast  disease.  Ultrasound 
Med.  &  Biol.  I,  393-404  (1975). 


Kobayashi ,  T. ,  Takatani,  0.,  Hattori ,  N., 
and  Kimura,  K. ,  Differential  diagnosis  of 
breast  tumors:    The  sensitivity  graded  method 
of  ul trasonotomography  and  clinical  evalua- 
tion of  its  diagnostic  accuracy.  Cancer  33, 
940-951  (1974). 

Pluygers,  E.,  Diagnostic  ultrasonore,  par 
echographie  A  et  B,  des  affections  Mammaires, 
J.  Beige.  Radiol.  58,  15-29  (1975). 

Wagai,  T.,  Tsutsumi ,  M. ,  and  Takeuchi,  H. , 
Diagnostic  Ultrasound  in  Breast  Diseases,  in 
Present  and  Future  of  Diagnostic  Ultrasound, 
I.  Donald  and  S.  Levi,  eds.,  pp.  148-161 

Okujima,  M.,  Refraction  of  ultrasonic  beam 
incident  near  circumference  of  spherical 
medium,  Proc.  Jap.  Soc.  Ultrasonics  Med.  29, 
231-232  (1976). 


[10]    Calderon,  C. ,  Vilkomerson,  D.,  Mezrich,  R. , 
Etzold,  K.  E.,  Kingsley,  B.,  and  Haskin,  M., 
Differences  in  the  attenuation  of  ultrasound 
by  normal,  benign  and  malignant  breast  tissue, 
J.  Clin.  Ultrasound  4,  249-254  (1976). 

[11]    Field,  S.  and  Dunn,  F.,  Correlation  of  echo- 
graphic  visualization  of  tissue  with  bio- 
logical composition  and  physiological  state, 
J.  Acoust.  Soc.  Am.  54,  809-812  (1973). 

[12]    Kobayashi,  T. ,  Takatani,  0.,  Hattori,  N.  , 
and  Kimura,  K. ,  Clinical  investigation  for 
the  differential  diagnosis  of  breast  tumor 
by  means  of  the  sensitivity  graded  method 
of  ul trasonotomography  (fourth  report)  - 
Analytical  re-appraisal  of  diagnostic 
criteria  in  echographie  and  mammographic 
findings,  Proc.  24th  Meeting  Jap.  Soc. 
Ultrasonics  Med.  24,  147-148  (1973). 


99 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


THE  ATTENUATION  OF  SELECTED  SOFT  TISSUE  AS  A  FUNCTION  OF  FREQUENCY 


D.  H.  Le  Croissette,  R.  C.  Heyser,  P.  M.  Gammell ,  and  J.  A.  Roseboro 

Jet  Propulsion  Laboratory 
California  Institute  of  Technology 
Pasadena,  California    91103,  U.S.A. 

and 

R.  L.  Wilsoni 

School  of  Medicine 
University  of  Southern  California 
Los  Angeles,  California    95033,  U.S.A. 


Measurements  of  attenuation  versus  frequency  have  been  made  on  pancreas,  kidney, 
fat  and  liver  specimens  of  hog  tissue.    Hogs  were  of  the  "Berkshire"  variety  and  the 
tissue  obtained  from  a  slaughterhouse.    The  measurements  were  taken  under  controlled 
conditions  of  temperature  and  formalin  fixing  to  determine  the  effects  of  aging  and 
fixing  on  the  tissue.    A  swept-f requency  transmission  system  was  used  operating  over 
the  range  of  about  1.5  to  9.5  MHz.    The  curves  show  the  changes  occurring  in  fresh 
tissue  as  a  result  of  several  days  of  aging  at  refrigerator  temperatures  (5  °C)  and 
the  effects  of  fixing  the  tissue  in  formalin  immediately  post-mortem  or  after  several 
days  of  5  °C  storage.    All  measurements  were  made  at  37  °C. 

Key  words:    Attenuation;  tissue  properties;  transmission;  ultrasound. 


1.  Introduction 

The  recent  interest  in  characterizing  soft  tis- 
sue by  ultrasound  is  based  upon  measurements  made 
as  early  as  1952  [2]^  and  later  observations  by 
scientific  investigators  attempting  to  lay  the 
groundwork  for  a  new  technique  of  ultrasonic  diag- 
nosis.   The  work  reported  here  is  the  start  of  a 
systematic  attempt  to  measure  the  attenuation  of 
soft  tissue  as  a  function  of  frequency.  This 
topic  was  chosen  because  of  the  belief,  substan- 
tiated by  initial  measurements  [2],  that  there  is 
I    a  variation  in  the  attenuation  of  ultrasound 

through  tissue  as  a  function  of  frequency  that  can 
be  correlated  with  the  pathological  state  of  the 
tissue.    The  measurements  reported  here  were  made 
j    on  excised  tissue  obtained  from  hogs.    This  paper 
I    reports  data  obtained  from  fresh,  aged  and  fixed 
specimens. 

In  a  typical  pulse-echo  ultrasound  imaging 
system,  the  frequency  variation  of  attenuation 
cannot  usually  be  identified.    Furthermore,  the 
ii    increase  in  attenuation  of  normal  tissue  as  a 
j    function  of  frequency  tends  to  mask  the  high 
1    frequency  behavior  of  tissue  when  a  measurement 
is  made  over  a  long  path  length,  i.e.,  far  from 
the  transducer.    These  two  factors  have  made  the 
identification  of  pathological  tissue  by  its 


^Current  address:    Harbor  General  Hospital, 

Torrance,  California  90509. 
^Figures  in  brackets  indicate  literature 

references  at  the  end  of  this  paper. 


frequency-dependent  characteristic  unlikely  and 
so  this  phenomenon  plays  little  part  in  the  sub- 
jective pattern  recognition  techniques  used  in 
contemporary  diagnostic  ultrasonography.    It  is 
believed,  however,  that  if  an  identification 
mechanism  based  on  frequency-dependent  phenomenon 
can  be  found,  suitable  instrumentation  will  be 
devised  for  tissue  identification. 

Measurements  on  tissues  of  known  pathologies 
can  greatly  increase  the  knowledge  of  tissue 
characterization  needed  to  develop  such  instru- 
mentation.   Specimens  obtained  from  cadavers 
are  especially  useful  since  the  individual  organ 
can  be  studied.    Most  autopsy  material,  however, 
is  not  available  until  after  a  one  or  two-day 
delay.    Furthermore,  the  general  availability  of 
pathological  tissue  is  greatly  enhanced  if  a 
study  is  not  constrained  to  specimens  a  few  hours 
old.    This  study  is  therefore  to  determine  the 
acoustic  attenuation  characteristics  of  several 
organs  as  a  function  of  time  after  death  and 
after  fixing  in  formalin  solution. 

2.  Methodology 

There  are  two  major  methods  for  examining  the 
ultrasonic  propagation  behavior  of  tissue  as  a 
function  of  frequency.    The  most  popular  method 
[3-5]  uses  the  frequency  spectrum  that  is  emitted 
by  a  heavily  damped  transducer  operated  in  a 
pulse  mode.    This  method  is  capable  of  giving 
useable  data  at  frequencies  several  times  the 
natural  resonant  frequency  of  the  crystal  with 
adequate  signal-to-noise  ratios.    By  contrast. 


101 


the  technique  used  in  this  work  does  not  pulse 
the  transducer  but  sweeps  the  operating  frequency 
over  a  wide  band.    This  technique  [6],  known  as 
Time  Delay  Spectrometry  (TDS),  reduces  the  ef- 
fects of  multipath  propagation  and  reverberation 
by  using  a  swept  frequency  transmitter  and  re- 
ceiver so  that  the  frequency  acts  as  a  "time 
tag"  to  the  signal.    The  sweep  of  the  receiver 
is  delayed  from  that  of  the  transmitter  so  it 
will  only  respond  to  the  signals  with  the  proper 
time  delay.    This  discriminates  against  signals 
with  different  arrival  times,  especially  those 
due  to  reverberation  and  multipath  propagation. 

The  data  were  taken  with  a  standard  swept-type 
spectrum  analyzer-tracking  generator  which  has 
been  modified  by  offsetting  the  tracking  generator 
by  a  constant  frequency.    A  Panoramic  SPA-3  spec- 
trum analyzer  and  Panoramic  G6  tracking  generator 
were  used.    This  tracking  generator  can  be  adjust- 
ed to  compensate  for  the  time  of  transit  of  the 
ultrasound  through  the  sample.    A  sweep  rate  of 
100  MHz'S"-'  and  3  dB  intermediate  frequency  band- 
width of  600  Hz  were  used.    If  the  sweep  frequency 
versus  time  of  the  spectrum  analyzer  is  linear, 
this  constant  frequency  offset  will  exactly  com- 
pensate for  a  constant  time  delay  of  the  ultra- 
sound through  the  specimen.    The  time  delay  will 
be  constant  across  the  frequency  sweep  if  both 
the  specimen  and  the  medium  in  which  it  is  im- 
mersed are  nondispersi ve.    Dispersion  of  the  me- 
dium and  the  sample  will  not  affect  the  measure- 
ments by  more  than  1  dB,  provided  the  average 
velocity  of  the  sample  and  medium  does  not  change 
by  more  than  1  percent.    Linearity  requirements 
on  the  sweep  of  the  spectrum  analyzer  are  ex- 
treme, however. 

The  measurements  were  made  over  the  frequency 
range  1.5  to  9.4  MHz.    Two  Aerotech  Alpha  trans- 
ducers were  used.    The  10  MHz,  0.63  cm  diameter 
(0.25  inches)  transducers  were  selected  so  that 
the  resonant  frequency  was  just  above  the  range 
of  interest.    This  frequency  was  chosen  so  that 
the  transducer  sensitivity  increased  with  fre- 
quency over  the  band.    This  compensates  for  the 
increase  in  tissue  attenuation  with  frequency  to 
give  a  more  nearly  constant  signal-to-noise  ratio 
over  the  selected  frequency  spectrum.    The  speci- 
men was  located  midway  between  the  transducers, 
which  were  12  cm  apart. 

3.    Specimen  Handl ing 

Specimens  of  tissue  from  hogs  were  studied  no 
more  than  four  hours  after  slaughter.    The  tissue 
type  and  the  time  sequence  of  the  measurements 
are  shown  in  figure  1.    All  specimens  were  heat 
sealed  in  a  plastic  bag  containing  either  0.9 
percent  saline  or  formaldehyde  diluted  10:1  with 
water  (i.e.,  approximately  4  percent).    The  liver 
specimens  were  sliced  to  approximately  15  mm 
thickness;  the  kidney  and  pancreas  specimens  were 
studied  intact.    The  backfat  specimen  was  left  in 
the  form  of  a  slab  as  it  was  stripped  from  the 
carcass.    The  thickness  of  all  of  the  specimens 
was  calipered  at  the  locations  studied.    Fat  is 
normally  removed  one  day  after  slaughter  which 
accounts  for  the  one-day  delay  shown  in  figure 
Kb). 

Figures  2  through  9  show  the  measured  attenua- 
tion versus  frequency  for  the  hog  tissue  at 
37  °C.    Four  or  five  locations  on  each  specimen 
were  studied.    Obvious  ducts,  vessels,  and  taper- 


DAYS  POST  MORTEM 

SPEC.  NO. 

0    1     1     1    2    1     3    1    4    1     5    1    6    1     7     1    8    1    9    1    10   1    11    1  12  1 

1  1  1 — .  —I  1  . — L — -  I  1  I  1  1  1  1  I 

PANCREAS  ^ 
KIDNEY 

r.  

I     205  9 

O  1      1      1      1      1  0  1  , . ,  .1 , ,    i,     lot      1      1  ;Sl 

1  2062 

1      1      1      1      1  O  )      lit      1      1      t  .^-^ 

1  2061 

1       1  O  1       1       1       t       1       )       1       1       )  1 

1  2060 

'  ■  .  l  ■        i   ■       1         t  .  . .  ,             .V-,',  ,  . .  ^.  .V-.  J^. -\-,  ,V        A  ^K\^\\\^ 

1  2047 

O  1       1  O  1       1       1  0  1,  .   f            :\X7.  }  t 

]  2064 

o  1     1  o  1     1     1  0  f     1     lot     I     t     t  ..^ 

1  2048 

1      1  O  1      }      t      1      1      J      1      I      1      i  ..  aN 

1  2049 

i    A     1     t     t     i     t     lot     1     1     t  §1 

1  2063 

1         1         1         1         1         t         t         t  O   t         t         1         1  NX\1 

1  2065 

11    1    1  |o|     1  oi,,,  t,„J„^,,t  

(a) 

h^AX'xv^NXKx^'^l  FIXED                          (:':-:^::v:-:';-:';''->v:'J  FROZEN 
1  1   5  °C                                   Q  DATA 

DAYS  POST  MORTEM 

SPEC.  NO. 

0|l|2|3l4l5|6|7|8|9llO|n|l2| 

FAT  - 

1  2055 

lot     1  o  1     1     1  oi-i?*:'  1    lot    1  1 

1  2057 

1     1     1     1    1    1  o'iixi'  t    1    1    1  1 

1  2056 

1     1     1  c  !    1     1    !     t    !    1    1     1  1 

1       4.:<i;.<.t.,.J,.-4.,,i<v.J,,                  J      A  1 

LI  m 

1  2050 

1  o  1     lo!     1     1  o^<^:V^^^^^^^^,;^t*^^^^^^^^^^  ,1 

1     206  7 

o  1    1  o  1     1     Id    1    t  o  1    1    1    1  ,';5l 

1  2052 

1     1     1     1     1  o  1    1    1     1    1    1  i 

I  2051 

1     1  o'l     r  !    1    1     i     t    1    1  1'^^ 

1  2Q66 

1   \t  '  1     r  t     1     1     i  c  1     1    i  15^ 

1  2053 

1     1     1     1     1     !     1     1  o  1     1     i,  ,  t_:^ 

1  2054 

1    t    t    1    1    f    1    1 o 1    i  ! 

1  2068 

1      1      i      1      1      i      1  O  i      1      1      1      t  OS) 

(b) 
Fig 

.  1. 

t"              J  FIXED                           [  :                1  FROZEN 
1                  1   5  °C                                    O  DATA 

Specimen  history:     (a)  pancreas  and 
kidney,  (b)  fat  and  liver. 

ed  edges  were  avoided.    In  the  case  of  the  kidney, 
measurements  were  not  made  in  the  central  portion, 
which  contains  the  large  vessels  and  the  renal 
pelvis.    All  of  the  kidney  measurements  were  made 
in  the  region  of  the  pyramids,  and  so  include 
both  cortical  and  medullary  matter.    The  loca- 
tions studied  on  any  given  specimen  can  be  iden- 
tified by  the  shape  of  the  symbol  used  on  figures 
2  through  5.    That  is,  the  circle,  square,  or 
triangle  represents  the  same  visible  location  on 
the  specimen  when  studied  on  the  day  of  slaughter, 
five  days  later,  and  after  fixing.    Care  was 
taken  to  avoid  multipath  effects  by  visual  in- 
spection of  the  area  of  the  specimen  and  by 
observing  the  spectrum  analyzer  for  obvious  signs 
of  multipath  interference.    Since  no  fine  struc- 
ture was  observed  in  the  attenuation  versus  fre- 
quency curves,  the  data  are  reported  at  0.5  MHz 
i  nterval s . 

Each  specimen  was  measured  at  four  or  five 
points.    This  is  a  compromise  between  sampling 
enough  locations  to  insure  that  the  measure- 
ments are  representative  and  reducing  experi- 
ment time  to  a  minimum.    Since  tissue  is  known 
to  deteriorate  rapidly  at  37  °C,  it  is  impor- 
tant that  the  experiment  time  be  kept  as  short 
as  possible.    Stray  air  bubbles  that  initially 
clung  to  the  specimen  were  gently  brushed  away. 
No  gas  evolution  was  subsequently  observed 
vi  sual ly. 


102 


2  4  6 

Frequency,  MHz 


10 


25r 


20- 


5  15- 


10- 


4  6 
Frequency,  MHz 


10 


25r 


20- 


15- 


10- 


25 


20 


15 


53 

u  lOh 

M- 


1  Oh  - 


20- 


I—  0, 


2  4  6  8 

Frequency,  MHz 


(d) 


A  A 
□ 

□ 

o 

O  V 


a  <s 

□  O 


□ 


Fig.  2. 


Attenuation  versus  frequency  for  hog  pancreas:  (a)  6  hours 
post-mortem  (fresh) ,  (c)  5  hours  post-mortem  (fixed),  (d)  5 


4  6  8 

Frequency,  MHz 

post-mortem  (fresh),  (b)  5  days 
days  post-mortem  (fixed). 


10 


These  measured  values  of  tissue  attenuation 
exhibit  considerable  variability.    At  this 
stage  in  the  experiments  it  was  considered  that 
little  was  to  be  gained  by  applying  statistical 
analysis  on  the  four  or  five  readings  taken  al- 
though it  is  recognized  that  a  mean  and  vari- 
ance will  be  meaningful  on  readings  from  a 
)  higher  number  of  points.    Accordingly,  the 
;  averaged  or  summarized  data  are  presented  in 
I  the  form  of  band  diagrams  in  figures  6  through 
J  9.    These  bands  were  selected  to  follow  the  data 
with  no  more  than  10  percent  of  the  points  fall- 
i  ing  outside  of  the  band. 


4.    Data  Acquisition 

The  response  of  the  system  was  recorded  with 
attenuations  of  0,  5,  10,  15,  20,  25,  and  30  dB 
switched  into  the  system.    The  response  with  the 
specimen  between  the  transducers  was  then  record- 
ed on  the  same  polaroid  film.    Since  the  sweep 
of  the  spectrum  analyzer  only  covers  a  5.5  MHz 
range  in  the  present  equipment,  it  was  necessary 
to  take  the  data  in  two  parts.    As  a  check, 
readings  taken  from  both  spectra  are  reported  in 
the  range  of  overlap,  which  is  4.5  to  5.5  MHz. 
If  only  one  reading  is  reported  at  a  given  loca- 


103 


OL-  0, 


4  6 
Frequency,  MHz 


2  4  6 

Frequency,  MHz 


25 


20 


S  15 


10 


25 


20 


15 


10 


OL 


2  4  6 

Frequency,  MHz 


10 


Frequency,  MHz 


Fig.  3. 


Attenuation  versus  frequency  for  hog  kidney:     (a)  3  hours  post-mortem  (fresh), 
post-mortem  (fresh) ,  (c)  4  hours  post-mortem  (fixed),  (d)  5  days  post-mortem  (f 


(b)  5  days 
i  xed ) . 


tion  on  the  specimen  then  the  readings  from  the 
upper  and  from  the  lower  spectra  agree  within 
0.5  dB.    Since  the  calibration  spectra  are  ob- 
tained for  5  dB  increments,  interpolation  to  the 
nearest  1  dB  is  reliable.    For  smoothness  in 
plotting  the  data,  the  graphs  are  plotted  to  the 
nearest  0.5  dB,  although  no  significance  should 
be  attached  to  changes  of  less  than  1  dB.  The 
accuracy  of  the  data  is  believed  to  be  generally 
±  1  dB,  with  rare  deviations  of  ±  2  dB.  Meas- 
urements using  standard,  non-biological  samples 
show  a  system  repeatability  error  of  +  0.5  dB. 


5.  Measurements 

Figure  2  shows  the  plot  of  attenuation  versus 
frequency  for  hog  pancreas  under  four  conditions; 
(a)  six  hours  post-mortem  (fresh),  (b)  five  days 
post-mortem  (fresh),  (c)  fixed  five  hours  post- 
mortem, and  (d)  fixed  five  days  post-mortem. 
Storage  of  the  fresh  tissue  at  5  °C  results  in  a 
decreased  attenuation  particularly  at  the  higher 
frequencies.    These  data  do  not  seem  conclusive, 
however,  since  figure  2(b)  shows  a  bimodal  dis- 
tribution of  readings.    The  organ  lacks  rigidity 
and  so  accurate  placement  and  replacement  of  the 


104 


25r 


20- 


15- 


10- 


25r- 


25|- 


20- 


15- 


15- 


."10- 


-      o  - 


10- 


o)  5 " 


OL-  OL 


2  4  6 

Frequency,  MHz 


4  6 
Frequency,  MHz 


8 

V  o 


A  □  i  o 

r\  o 

O  g  8  ° 


□  □  O 


25 


20- 


15 


10 


2  4  6 

Frequency,  MHz 


10 


OL-  0, 


2  4  6 

Frequency,  MHz 


10 


Fig.  4. 


Attenuation  versus  frequency  for  hog  backfat:  (a)  1  day  post-mortem  (fresh),  (b)  6  days 
post-mortem  (fresh) ,  (c)  1  day  post-mortem  (fixed,  (d)  6  days  post-mortem  (fixed). 


transducers  on  these  specimens  is  very  difficult. 
In  addition,  the  pancreas  contains  lobules  of 
glandular  tissue  on  the  order  of  1  cm,  inter- 
spersed with  fat.    It  is  therefore  considered 
that  the  wide  spread  in  data  can  be  accounted  for 
by  the  lack  of  rigidity  and  the  inhomogeneity  of 
the  specimen.    Figure  6  shows  the  bands  of  atten- 
uation versus  frequency  for  the  fresh  specimens 
under  two  conditions;  six  hours  and  five  days 
post-mortem. 

The  measurements  on  fresh  and  fixed  kidneys 
are  shown  in  figure  3  and  the  corresponding  bands 
of  attenuation  are  given  in  figure  7.    For  the 


fresh  specimens,  storage  of  the  tissue  at  5  °C 
for  five  days  has  resulted  in  a  decrease  in  at- 
tenuation, especially  at  the  higher  frequencies. 
Measurements  taken  48  hours  post-mortem  (not  re- 
ported here)  lie  between  the  two  values.  This 
decrease  is  just  sufficient  to  separate  the  band 
curves.    The  attenuation  of  fixed  tissue  is 
above  that  of  the  tissue  just  post-mortem. 

Figures  4  and  8  made  on  hog  fat  show  that  there 
is  little  effect  of  aging  on  the  attenuation  char- 
acteristics. Little  significance  can  be  placed  on 
the  differences  at  low  frequencies  shown  in  figure 
8  since  the  attenuation  is  only  of  the  order  of 


105 


25 


25 


15- 


sio- 


(a) 


4  6 
Frequency,  MHz 


20 


15 


10 


10 


15 


10 


(b) 


X 


^  a 


C)  □ 


2  4 

Frequency, 


6 

MHz 


10 


251- 


20 


15 


10 


OL- 


Frequency,  MHz 


4  6 
Frequency,  MHz 


Fig.  5. 


Attenuation  versus  frequency  for  hog  liver:  (a)  5  hours  post-mortem  (fresh),  (b)  5  days 
post-mortem  (fresh) ,  (c)  5  hours  post-mortem  (fixed),  (d)  5  days  post-mortem  (fixed). 


the  uncertainties  in  the  measurements.    Fixing  the 
specimens  gives  a  slightly  higher  attenuation. 
There  is  considerable  difference  between  this  fat 
and  typical  fat  found  in  human  studies  so  that 
these  data  should  be  applied  with  caution.  Hog 
fat  is  white  and  of  a  uniform  texture  whereas 
human  fat  is  yellow  and  has  a  lobular  structure 
with  connective  tissue  interspersed. 

Figures  5  and  9  show  the  data  for  hog  liver. 
Here  the  storage  of  a  fresh  specimen  results  in 
a  lower  attenuation  over  a  five-day  period.  The 
fixed  specimens  show  increased  attenuation.  The 
data  bands  are  fairly  narrow  probably  because  of 


the  homogeneity  of  the  tissue  at  the  resolution 
level  studied. 

These  tissue  specimens  were  stored  at  5  °C  and 
heated  to  the  measurement  temperature  of  37  °C 
over  about  a  one-hour  period.    Following  this 
procedure,  each  specimen  was  returned  to  the  re- 
frigerator for  storage  until  the  next  measurement 
was  made  or  until  the  specimen  was  fixed.  For 
comparison,  some  specimens  were  held  undisturbed 
at  5  °C  until  a  single  measurement  was  taken.  No 
significant  difference  in  attenuation  could  be 
seen  between  the  two  groups. 


106 


T 


6  h  post-mortem 
5  d  post-mortem 


4  6 
Frequency,  MHz 


10 


Attenuation  versus  frequency  data  bands 
for  hog  pancreas:    6  hours  and  5  days 
post-mortem  (fresh). 


5 


3  h  post-mortem 
iyZ\  5  d  post-mortem 


4  6 
Frequency,  MHz 


10 


Fig.  7, 


Attenuation  versus  frequency  data  bands 
for  hog  kidney:    3  hours  and  5  days 
post-mortem  (fresh). 


T 


Frequency,  MHz 

8.    Attenuation  versus  frequency  data  bands 
for  hog  backfat:     1  day  and  6  days 
post-mortem  (fresh). 


T 


5  h  post-mortem 
5  d  post-mortem 


Frequency,  MHz 

Fig.  9.    Attenuation  versus  frequency  data  bands 
for  hog  liver:    5  hours  and  5  days 
post-mortem  (fresh). 


107 


6.  Discussion 

The  accuracy  of  these  measurements  relies 
upon  the  internal  sweep  linearity  of  the  spectrum 
analyzer  to  provide  a  constant  time  delay  across 
the  frequency  band  matching  the  constant  transit 
time  of  the  sound  energy  through  the  specimen. 
Slight  non-linearities  in  the  sweep  were  observed 
to  give  an  error  as  high  as  1  dB  in  the  measured 
attenuation.    Some  indication  of  this  error  may 
be  seen  in  the  region  between  4.5  and  5.5  MHz. 
In  practice,  two  sets  of  data  were  taken;  the 
lower  half  extended  from  1.5  to  5.5  MHz  and  the 
upper  part  from  4.5  to  9.5  MHz.    Small  errors  in 
the  overlapping  region  are  indicated  by  two  val- 
ues being  plotted  with  the  same  symbol.    In  many 
cases  identical  values  were  obtained  or  the  er- 
ror was  small.    Since  these  measurements  were 
taken,  a  new  system  has  been  built  in  which  the 
sweeps  of  the  transmitter  and  receiver  are  ob- 
tained from  digitally  programmed  phase-coherent 
synthesizers.    This  will  insure  that  the  system 
is  aligned  over  the  entire  frequency  range  of  1 
to  10  MHz  and  will  materially  improve  the  quality 
of  the  data . 

Acknowledgment 

This  paper  presents  the  results  of  one  phase 
of  research  conducted  at  the  Jet  Propulsion 
Laboratory,  California  Institute  of  Technology 
for  the  National  Science  Foundation,  by  agree- 
ment with  the  National  Aeronautics  and  Space 
Administration. 

References 


[2]    Le  Croissette,  D.  H.  and  Heyser,  R.  C, 
Attenuation  and  Velocity  Measurements  in 
Tissue  Using  Time  Delay  Spectrometry,  in 
Ultrasonic  Tissue  Characterization,  M.Linzer, 
ed..  National  Bureau  of  Standards  Spec.  Publ. 
453,  pp.  167-196  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

[3]    Lele,  P.  P.,  Mansfield,  A.  B.,  Murphy,  A.  I., 
Namery,  J.,  and  Senapati ,  N.,  Tissue  Charac- 
terization by  Ultrasonic  Frequency-Dependent 
Attentuation  and  Scattering,  in  Ul trasoni c 
Tissue  Characterization,  M.  Linzer,  ed.. 
National  Bureau  of  Standards  Spec.  Publ.  453, 
pp.  167-196  (U.  S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

[4]    Lizzi,  F.  L.  and  Laviola,  M.  A.,  Tissue 

signature  characterization  using  frequency 
domain  analysis.  Ultrasonics  Symposium 
Proceedings  (IEEE),  pp.  714-719,  1976. 

[5]    Miller,  J.  G.,  Yuhas,  D.  E.,  Mimbs,  J.  W. , 
Dierker,  S.  B.,  Buse,  L.  J.,  Laterra,  J.  J., 
Weiss,  A.  N. ,  and  Sobel ,  B.  E.,  Ultrasonic 
tissue  characterization:    correlation  be- 
tween biochemical  and  ultrasonic  indices  of 
myocardial  injury.  Ultrasonics  Symposium 
Proceedings  (IEEE),  pp.  33-43,  1976. 

[6]    Heyser,  R.  C.  and  Le  Croissette,  D.  H., 
A  new  ultrasonic  imaging  system  using  Time 
Delay  Spectrometry,  Ultrasound  in  Med,  and 
Biol .  1,  119-131  (1974). 


[1]    Wild,  J.  J.  and  Reid,  J.  M.,  Further  pilot 
echographic  studies  on  the  histologic  struc- 
ture of  tumor  of  the  living  and  intact  human 
breast.  Am.  J.  Pathol.  28,  831-861  (1952). 


108 


CHAPTER  4 
SCATTERING  AND  ATTENUATION 


109 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


CLINICAL  SPECTRUM  ANALYSIS  TECHNIQUES  FOR  TISSUE  CHARACTERIZATION 


Frederic  L.  Lizzi  and  Marek  E.  Elbaum 

Riverside  Research  Institute 
80  West  End  Avenue 
New  York,  New  York    10023,  U.S.A. 


A  spectrum  analysis  system  is  being  used  in  a  clinical  measurement  program  to  ex- 
amine the  reflectance  characteristics  of  intraocular  structures.    The  on-line  system 
computes  power  spectra  descriptive  of  echoes  returned  from  well-defined  tissue  seg- 
ments chosen  by  the  examiner.    Clinical  data  are  digitized  for  subsequent  post- 
processing and  cataloguing.    This  paper  presents  an  overview  of  the  theoretical 
analysis  used  to  formulate  system  design  and  to  guide  data  interpretation.  The 
analysis  accounts  for  stochastic  tissue  architectures  and  treats  the  specific  il- 
lumination conditions  employed  in  the  system.    The  paper  also  discusses  system  param- 
eters as  related  to  observable  tissue  characteristics  and  presents  typical  test 
data  and  clinical  spectra  for  several  pathologic  conditions. 

Keywords:    Clinical  ultrasound;  ocular  tumor;  power  spectra;  Rayleigh  scattering; 
ultrasonic  spectrum  analysis. 


1.  Introduction 

Within  the  past  few  years,  a  variety  of  ad- 
vanced techniques  have  been  employed  to  measure 
the  ultrasonic  properties  of  soft-tissue  struc- 
tures [l-3]i.    This  paper  describes  a  clinical 
spectrum  analysis  technique  [4]  which  has  been 
used  to  measure  the  frequency-domain  charac- 
teristics of  ultrasonic  backscatter  from  tissues 
of  the  eye  and  orbit.    The  technique  employs  a 
clinical,  on-line  system  together  with  computer 
post-processing  and  data  cataloguing.  Following 
laboratory  verification,  an  extensive  series  of 
clinical  measurements  has  been  carried  out.  The 
in  vivo  spectral  data  base  is  now  being  evalu- 
ated in  terms  of  potential  diagnostic  signifi- 
cance and  is  also  being  employed  in  analytic  model- 
ling of  tissue  reflectance. 

The  spectrum  analysis  techniques  have  been  de- 
signed to  accommodate  the  stochastic  nature  of 
clinically  relevant  tissue  structures,  such  as 
ocular  tumors  and  vitreous  hemorrhages.  These 
structures  exhibit  randomness  in  terms  of  the  size, 
orientation,  and/or  spatial  position  of  their  con- 
stituent internal  scattering  centers.    Single  re- 
flectance measurements  made  on  such  stochastic 
entities  exhibit  a  significant  degree  of  statisti- 
cal fluctuation  which  can  preclude  the  reliable 
identification  of  pertinent  tissue  characteristics. 
Accordingly,  the  spectrum  analysis  techniques  de- 
scribed below  incorporate  on-line  ensemble  averag- 
ing:   the  resulting  power  spectra  determinations 
constitute  statistically  stable  descriptors  of  tis- 
sue properties.    In  addition,  compensatory  proce- 
dures are  employed  to  account  for  extraneous  spec- 
tral weighting  introduced  by  the  transfer  functions 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


of  transducers  and  electronic  subsystems. 

The  spectral  techniques  can  ultimately  be  em- 
ployed clinically  to  examine  organs  other  than  the 
eye.    However,  their  in  vivo  application  is  now 
practicable  in  ocular  examinations  since  the  media 
of  the  eye  do  not  present  additional  constraints 
found  in  other  organs.    If  transmission  through 
the  ocular  lens  is  avoided  [5],  acoustic  attenua- 
tion and  refraction  do  not  present  significant 
obstacles  in  viewing  intra-ocular  structures  at 
center  frequencies  higher  than  10  MHz.    The  large 
bandwidths  that  can  be  realized  at  these  high  fre- 
quencies permit  spectral  sensitivity  to  morpho- 
logical features  whose  scale  sizes  are  significant- 
ly less  than  0.5  mm.    In  addition,  the  narrow  beam- 
widths  (e.g.  ,  0.3  mm)  available  at  these  center 
frequencies  provide  precise  spatial  definition  of 
tissue  segments  to  be  analyzed  and  permit  meaning- 
ful ensemble  averaging,  even  within  small  tissue 
vol umes . 

To  date  well  over  200  clinically  derived  spectra 
of  normal  and  abnormal  structures  have  been  ac- 
quired, processed,  and  catalogued  in  disease- 
indexed  digital  files.    Previous  publications  have 
discussed  the  operation  of  the  spectrum  analysis 
system  and  presented  clinical  findings  [6,7]. 
This  paper  first  outlines  the  analytic  framework 
which  has  guided  system  design,  clinical  operation, 
and  data  interpretation.    It  then  describes  system 
operation  emphasizing  the  relations  between  system 
performance  parameters  and  observable  tissue  charac- 
teristics.   Lastly,  it  presents  representative  test 
results  for  comparison  with  clinical  data  obtained 
for  several  ocular  pathologies. 

2.    Outline  of  Analytic  Framework 

The  clinical  deployment  of  the  spectrum  analysis 
techniques  employed  in  the  present  system  has  been 


111 


guided  by  an  analysis  which  accounts  for  both  the 
stochastic  nature  of  tissue  architectures  and  the 
geometry  and  temporal  spectrum  of  the  ultrasonic 
illumination  beam.  This  analysis  is  presented  in 
outline  form  to  facilitate  discussions  in  follow- 
ing sections.  (A  more  comprehensive  treatment  will 
be  presented  elsewhere  [8].) 

The  analysis  is  discussed,  first,  in  terms  of 
its  general  formulation  which  accounts  for  the  use 
of  focused  illumination  conditions.    Second,  tis- 
sue architectures  which  are  statistically  station- 
ary are  discussed.    Lastly,  attention  is  given  to 
tissues  which  can  be  modelled  as  randomly  distribut- 
ed Rayleigh  scatterers. 

A.    Basic  Formulation 

The  geometry  involved  in  the  analysis  is  shown 
in  figure  1.    A  focused  transducer  emits  a  short, 
broadband  ultrasonic  pressure  pulse.    Radio  fre- 
quency echo  signals  received  from  scatterers  within 
a  range-gated  segment  (0-L)  are  subjected  to  spec- 
trum analysis.    The  range  gate  is  positioned  within 
the  focal  volume  of  the  transducer  and  its  time 
duration  is  chosen  to  be  much  larger  than  the  dura- 
tion of  echoes  from  a  single  scatterer.    The  trans- 
mission characteristics  of  all  intervening  struc- 
tures are  assumed  to  be  statistically  homogeneous 
and  to  result  in  negligible  attenuation. 


X 


Gated 
Segment 

Fig.  1.    Geometry  for  spectrum  analysis 
measurements. 


widths  (e.g.  ,  0.3  degree)  and  small  fractional 
range  segments,  L/R,  that  are  of  interest,  the 
beam  pattern  over  the  gated  segment  can  be  expres- 
sed as  a  function  of  cross-sectional  arguments 
only:    i.e.,  F(re)  =  F(y,z).    Also,  the  range,  r, 
can  be  replaced  by  R  +  x.    Thus,  eq.  (1)  can  be 
rewritten  as 

„-jwR/c     ■  , 

p'(co)  =  (jcoA/c)P(co)^  e  J"''/''F(y,z)  .  (2) 

2ttR 

The  backscatter  from  the  gated  tissue  volume  is 
characterized  in  terms  of  a  scattering  function, 
S(a);  x,y,z).    This  function  relates  the  amplitude 
of  the  incident  pressure  at  a  point  (x,y,z)  within 
the  volume  of  interest  to  the  strength  of  the  back- 
scattered  spherical  wave  originating  at  that  point. 
The  backscattered  wave  is  sensed  by  the  transducer 
which  responds  to  the  spatial  average  of  the  pres- 
sure distribution  generated  over  its  surface.  The 
spatially  averaged  pressure  received  from  each 
scattering  element  is  proportional  to  F2(y,z)  and 
can  be  readily  calculated  by  using  the  coordinate 
system  employed  by  O'Neil  [9].    The  total  received 
pressure  is  then  obtained  by  integration  over  the 
gated  volume,  v,^ 

p,^(co)  =  B(a,)yS(a3;  x,y,z)F2(y,z)e-j2'^^/^dxdydz  (3a) 
^m 

where  B(u))  is 

^-j2coR/c 

B(aj)  =  (jwA/c)P(oj)^   (3b) 

(2TrR)2 


and  the  sub-index  m  indicates  the  m'-''  measurement. 

As  a  first  approximation  for  many  tissues  of 
interest,  S(u);  x,y,z)  characterizes  weak  scatter- 
ing of  identical,  randomly  distributed,  isotropic 
scattering  elements.    Under  these  conditions  it 
is  convenient  to  factor  the  scattering  function 
as  follows 


The  received  signals  are  analyzed  in  the  fre- 
quency domain.    The  classic  results  obtained  by 
O'Neil  [9]  are  applicable  for  the  weakly  focused 
transducers  used  in  this  program.    These  results 
show  that,  at  a  point  Q  in  the  focal  zone,  the  in- 
cident wave  is  essentially  planar  and  its  pressure 
amplitude  is  given  by 


p-jkr 

p'  (oj)  =  jkA  P{m)-  F(r,e) 

2TTr 


(1) 


where  k  =  u/c  is  the  wave  number;  u  and  c  are 
radian  temporal  frequency  and  propagation  velocity, 
respectively;  P(a))  is  the  pressure  amplitude  at  the 
transducer  surface  with  area  A.    The  function 
F(r,e)  describes  the  beam  pattern  at  a  constant 
range  from  the  transducer  and  is  equal  to 
2Ji(k  a  sin  e)/(k  a  sin  e)  where  Ji(6)  represents 
a  Bessel  function  of  the  first  kind  and  first 
order  with  argument  g. 

It  is  useful  to  simplify  this  expression  at  the 
outset.    By  virtue  of  the  narrow  angular  beam- 


S(w;  x,y,z)  =  H(a))N(x,y,z) 


(4) 


Here  H(aj)  defines  the  reflectance  characteristic 
of  a  single  scattering  element  while  N(x,y,z) 
describes  their  effective  spatial  concentration. 
With  eqs.  (3)  and  (4),  the  total  received  pres- 
sure in  the  m^'^  measurement  can  be  written  as  a 
Fourier  transformation,  with  respect  to  axial  range, 
of  the  product  of  the  gating  function  G(x)  and  a 
function  characterizing  the  weighted  concentration 
of  tissue  elements;  i.e. 


p^(w)  =  B(a))H(w) J  G(x)|" 


'N(x,y,z; 


(5) 


F2(y,z)dydz 


e-j2(^x/c^^ 


The  function  G(x)  describes  the  length  of  the  range 
gate  and  the  range-weighting  function  (e.g. ,  Ham- 


112 


ming)  applied  in  the  measurement.    Denoting  the 
Fourier  Integral  by  Qfji(co),  eq.  (5)  can  be  rewritten 
as 


p  (w)  =  B(a))H(a))Q^((. 
'^m  m 


(6) 


It  is  important  to  notice  the  stochastic  nature 
of  Qm(a))  resulting  from  the  stochastic  nature  of 
the  tissue  concentration  function,  N(x,y,z).  Quali- 
tatively, for  equal,  but  non-overlapping  tissue 
volumes,  Pm(u)  can  vary  randomly  from  measurement 
to  measurement,  as  different  tissue  realizations 
are  illuminated  with  the  ultrasonic  beam.  This 
stochastic  process  can  be  meaningfully  character- 
ized in  terms  of  the  power  spectrum,  p,  computed 
by  forming  an  ensemble  average  of  the  modulus- 
square  of  Pni(a))  -,  i  ■  e. , 


<N(x,y,z)N(xi,yi,Zi)>  =  R.,(Ax,Ay,Az) 


(9) 


where  Ax  =  x-xj.  Ay  =  y-yi,  az  =  z-zi.  For  this 
case,  substitution  into  eqs.  (5)  through  (7)  can 
be  carried  out  with  considerable  simplification. 
The  expected  value  of  p,  taken  in  the  sense  of  an 
average  over  the  ensemble  of  tissue  realizations 
becomes 


<P> 


(10) 


|B(aj)|: 


|H(w)|2y  J  J  Rg(Ax)  Rp(Ay,Az) 


P  =  1  E  |P  (f^) 
M        ' ^m 
m=l 


(7a) 


(7b) 


m=l 


In  the  present  technique,  the  ensemble  average 
is  computed  from  a  series  of  observations  of 
|Pi,i(oj)[2  taken  along  adjacent,  non-overlapping 
beam  headings  within  the  examined  tissue.    The  re- 
sulting function  p  then  constitutes  an  estimate  of 
the  "true"  power  spectrum,  <p>.     (The  symbol  <•> 
refers  to  ensemble  averaging  over  all  realizations 
of  the  structure  of  interest.) 

The  statistical  fluctuations  associated  with 
estimates  of  <p>  are  of  significant  practical  im- 
portance.   For  tissues  characterized  by  a 
Gaussian-distributed  concentration  function  and 
a  correlation  volume  much  smaller  than  the  sampling 
volume,  Vpi,  Pni(a))  has  a  Gaussian  distribution  and 
|Pni(u)p  obeys  Chi-square  statistics.  (Tissues 
which  possess  many  independent  scattering  centers 
within  Vm  obey  these  statistics  as  well  by  virtue 
of  the  law  of  large  numbers.) 

It  follows  from  the  properties  of  the  Chi-square 
distribution  that  independent  measurements  of 
|P[ii(a))|2  exhibit  significant  variability;  in  fact, 
at  any  value  of  u,  the  standard  deviation,  a,  of 
such  measurements  is  equal  to  their  mean  value, 
<p>.    The  relative  fluctuations  which  persist  after 
averaging  M  independent  realizations  of  |Pu,(a))|2 
can  be  assessed  from  the  ratio 


<P>  ±  o 
<P> 


=  1  ±  M" 


18) 


In  spectral  measurements,  this  degree  of  fluctua- 
tion is  evidenced  at  each  value  of  u  producing  a 
statistical  "ripple"  of  the  above  magnitude  about 
the  mean  spectral  shape.    This  result  has  been 
employed  to  select  a  value  of  M  that  is  suitable 
for  clinical  measurements  as  discussed  below. 

B.    Statistically  Stationary  Architectures 

If  the  concentration  function,  N(x,y,z),  is  a 
weakly  stationary  process,  then  its  autocorrela- 
tion function  can  be  expressed  as 


R^(Ax,Ay,Az)  x  ^-^^^^^/^  d(Ax)d(Ay)d(Az) 


where  Rq  and  Rp  are  the  autocorrelation  functions 
of  the  gate  function,  G,  and  directivity  function, 
F^,  respectively. 

For  a  limiting  case  that  is  of  practical  in- 
terest here,  the  correlation  volume  defined  by  the 
system-related  functions  Rq  and  Rp  is  much  larger 
than  that  of  the  tissue  function  Rn.  Equation 
(10)  can  then  be  approximated  as 


<P> 


^  |B(a))|2  |H(w)|2  Rg(0)  Rp(0,0)  T(a3)  (11) 


where  T(aj)  depends  only  on  the  tissue  structure 
and  is  given  by 


T(w)  ^  J  f  f  Rf^(Ax,Ay,Az)  d(Ay)d(Az 


(12) 


^-j2a.Ax/c 


C.    Spatial  Distribution 
of  Rayleigh  Scat<-erers 

The  above  discussions  are  applicable  directly 
to  tissues  which  can  be  modelled  as  a  set  of  ran- 
domly distributed,  discrete  Rayleigh  scatterers: 
i.e.,  scatterers  whose  sizes  are  much  smaller  than 
the  illumination  wavelengths.    We  next  treat  the 
case  where  the  number  of  Rayleigh  scatterers  with- 
in a  unit  volume  is  a  Poisson  process;  i.e., 


K 


N(x,y,z) 


^  6(x-x.)  6(y-y.)  6(z-z.: 
i  =  l 


(13) 


where  6  is  the  Dirac  delta  function  and  (xi,yi,z-j) 
are  the  random  coordinates  of  the  ith  scatterer 
which  are  uniformly  distributed  over  the  inspected 
tissue  volume.    K  is  the  number  of  scatterers  per 
unit  volume  and  obeys  Poisson  statistics  with  para- 
meter <K>.    It  follows  [10]  that  R^,  the  autocor- 
relation function  of  N,  is 


=  <K>  6(Ax)  6(Ay)  6(Az)  +  <K>2 


(14) 


113 


Using  eq.  (14)  for       in  eq.  (10),  one  obtains 


<P>  =  |B(a))|2  |H(to)|2    <K>Rg(0)Rp(0,0)  +  (15) 


<K>2  |G(a))  |2  /    /    Rj-(Ay,Az)  d(Ay)d(Az) 


where  G(u))  is  the  Fourier  transform  of  Rg(ax). 

The  first  term  in  eq.  (15),  proportional  to 
the  expected  number  of  scatterers,  is  identi- 
fied with  non-coherent  scattering.    The  second 
term,  proportional  to  the  square  of  the  ex- 
pected number  of  scatterers,  is  identified 
with  coherent  scattering.    The  relative  con- 
tributions of  both  components  to  the  total  mea- 
surement varies  with  frequency.    The  non- 
coherent spectral  component  is  distributed  over 
the  frequency  domain,  whereas  the  coherent  com- 
ponent primarily  contributes  at  zero  frequency 
providing  that  the  power  spectrum  |G(u))|2  can 
be  adequately  approximated  as  a  Dirac  delta 
function  at  the  frequency  origin. 

In  practice,  non-coherent  spectra  predominate 
in  clinical  observations.    This  occurs  because 
spectral  observations  are  carried  out  at  high 
frequencies  and  a  time-weighting  function  is  used 
to  suppress  the  sidelobes  of  |G(a))|2.    These  side- 
lobes  fall-off  at  a  rate  which  is  more  rapid  than 
the  a)"2  characteristic  associated  with  a  rectangu- 
lar gating  function. 

Equation  (15)  can  be  used  to  address  the  topic 
of  spectral  normalization.    The  non-coherent  term 
is  proportional  to  |H(a3)|2  which  for  Rayleigh  scat- 
terers has  an  aj'+  dependence.    However,  other 
system-related  terms  are  also  frequency-dependent. 
To  account  for  these  factors,  spectral  data  are 
normalized  using  calibration  spectra  obtained  from 
a  flat,  water-glass  interface  viewed  at  normal  in- 
cidence and  located  at  R  +  L/2.    It  can  be  shown 
that  this  normalization  removes  the  system-related 
frequency  dependence  in  the  non-coherent  component 


of  eq.  (15)  and  that  the  normalized  spectrum  is 
indeed  proportional  to  u'*.    Experimental  results 
verifying  this  conclusion  are  presented  in  a  sub- 
sequent section. 

These  results  and  those  of  preceding  sections 
are  applicable  to  a  variety  of  tissue  structures. 
However,  all  tissues  of  interest  do  not  fall  with- 
in simple  categories  and  further  analysis  is  war- 
ranted to  treat  more  complex  tissue  structures 
and  to  investigate  the  most  appropriate  approaches 
to  data  normalization.    Such  analyses  are  being 
conducted  as  the  clinical  data  base  is  expanded 
and  histologic  preparations  become  available. 

3.    Spectrum  Analysis  System 

The  spectrum  analysis  techniques  described  above 
have  been  implemented  with  an  on-line  clinical  sys- 
tem.   The  system  is  integrated  with  a  high- 
resolution  A-  and  B-scan  instrument  which  is  em- 
ployed routinely  in  ophthalmic  examinations  (fig. 
2).    The  operation  of  the  system  is  described 
elsewhere  [6]  and  is  only  briefly  summarized  here. 
The  tissue  segment  to  be  analyzed  is  selected  by 
observing  A-  and  B-scan  displays  in  which  the  posi- 
tion of  the  moveable,  system  range  gate  is  super- 
imposed.   These  displays  are  also  monitored  to  in- 
sure that  the  beam  does  not  traverse  the  absorp- 
tive ocular  lens  and  that  the  tissue  segment  to  be 
analyzed  lies  within  the  transducer's  focal  zone. 

After  the  range  gate  has  been  placed  over  the 
desired  tissue  segment,  processing  is  initiated. 
First,  the  gated  rf  echo  complex  is  multiplied  by 
a  time-weighting  function  to  suppress  spectral  side 
lobes  which  would  prevent  accurate  measurements  of 
steeply  rising  or  falling  spectra.    Then,  the 
gated  signals  are  applied  to  a  scanning  electronic 
spectrum  analyzer  which  computes  the  desired  spec- 
trum comprised  of  50  spectral  elements  occupying 
adjacent  300-KHz  frequency  bands. 

An  ensemble  of  these  spectra  are  computed  along 
13  adjacent,  non-overlapping  beam  headings  and 
entered  into  an  on-line  averager.    The  spectral 
ensemble  is  obtained  by  executing  a  slow  sector 
scan  during  which  an  optical  encoder  initiates  a 
spectrum  analysis  computation  each  time  the  trans- 


ELECTRONIC 
PULSER 


T/R 

TIME-WEIGHTED 

SCANNING 

SPECTRUM 

RANGE  GATE 

— a» 

SPECTRUM 

SWITCH 

UNIT 

ANALYZER 

AVERAGER 

SECTOR  SCAN 


ANALYSIS  REGION 
(1.5  X  4  0  mm  ) 


A/B -SCAN 
MONI  TORS 


STRI  P 
CHART 
RECORDER 


Fig.  2.    Configuration  of  spectrum  analysis  system. 


114 


ducer  heading  is  incremented  by  one  angular  beam- 
width  (0.3  degrees  at  10  MHz).     (In  practice,  100 
partly  redundant  spectra  are  averaged  to  improve 
electronic  signal-to-noise  ratios.)  Averaged 
spectra  are  presented  on  a  strip  chart  recorder 
which  also  displays  frequency  and  amplitude  cali- 
bration signals  for  use  in  subsequent  processing. 

After  each  examination  session,  calibration 
spectra  are  recorded  from  an  optically  flat  glass 
plate  situated  at  each  of  the  ranges  used  in  tis- 
sue measurements.    The  calibration  spectra  and 
tissue  spectra  are  subsequently  digitized  and 
entered  into  a  computer  for  spectral  normalization, 
smoothing,  regression  analysis,  and  data  catalogu- 
ing. 

Table  1  lists  the  salient  system  parameters 
which  influence  the  detail  with  which  tissue  char- 
acteristics such  as  |H([jj)|2  and  T(aj)  can  be  examin- 
ed.   As  discussed  below,  the  frequency  coverage, 
spectral  resolution,  and  number  of  non-overlapping 
beam  positions  are  of  central  importance  in  tissue 
studies. 

Table  1.    Nominal  system  parameters. 
Frequency  coverage  5  to  13  MHz 


Spectral  resolution 
Gating  function 
Spectrum  analyzer 

Gated  range  segment 

Lateral  scan  dimension 

Beam  width  (10  MHz) 


0.6  MHz 
0.3  MHz 

1. 5  mm 

4  mm 

0. 3  mm 


Number  of  distinct  beam  positions  13 
Dynamic  range  30  to  35 


The  frequency  coverage  achievable  in  spectral 
measurements  has  been  found  to  approximate  the 
20-dB  system  bandwidth  as  defined  by  glass  plate 
spectra.    This  coverage  is  nominally  5  to  13  MHz 
for  a  broadband  transducer  with  a  10  MHz  resonant 
frequency.    Beyond  this  range,  limited  signal-to- 
noise  ratios  are  encountered  and  spectral  normali- 
zation can  become  inaccurate.    The  available  fre- 
quency coverage  determines  the  spectral  charac- 
teristics of  tissue  elements  with  specific  scale 
sizes.    Within  the  above  frequency  range,  Rayleigh 
scattering  is  encountered  for  spherical  elements 
whose  diameters  are  less  than  35  ym.    As  discussed 
below,  membranes,  or  periodic  features,  with  axial 
dimensions  larger  than  100  pm  will  produce  period- 
ic spectra  with  at  least  one  spectral  cycle  dis- 
played over  the  achievable  8  MHz  coverage.    In  ad- 
dition, tissue  septa  whose  thicknesses  are  less 
than  30  ym  will  exhibit  monotonical ly  rising 
spectra. 

Spectral  resolution  is  determined  by  the  2-us 
gate  duration  and  the  weighting  function.  These 
variables  have  been  chosen  to  achieve  a  0.6  MHz 
resolution  while  the  spectrum  analyzer  employs  a 
0.3  MHz  filter  bandwidth.    The  short  gate  duration 
corresponds  to  a  1.5  mm  tissue  depth  permitting 
analysis  within  small  ocular  tumors  and  avoiding 
significant  signal  weighting  due  to  attenuation 
within  the  gated  echo  complex. 


The  number  of  independent  spectra  which  are 
ensemble  averaged  is  an  important  factor  in  assess- 
ing the  residual  statistical  fluctuation  with 
estimations  of  mean  spectral  shape  and  amplitude. 
Usually,  13  spectra  from  non-overlapping  beam 
positions  are  used  to  form  the  ensemble  average. 
For  the  conditions  described  with  reference  to  eq. 
(8),  a  statistical  spectral  "ripple"  of  approxi- 
mately ±  1  dB  is  expected.    While  spectral  fluc- 
tuations are  often  confined  within  this  range, 
there  are  situations  where  significantly  larger  ex- 
cursions occur  suggesting  that  the  analyzed  tis- 
sues exhibit  relatively  large  correlation  volumes. 

(Further  analyses  of  these  parameters  in  terms 
of  the  accuracies  of  spectral  slope  and  attenuation 
estimations  have  been  presented  in  a  previous 
publ ication  [7] . ) 


4.    Representative  Results 

Before  clinical  deployment,  the  performance  of 
the  spectrum  analysis  system  and  the  applicability 
of  the  theoretical  approach  was  verified  on  a  num- 
ber of  test  targets.    One  type  of  test  target  was 
a  thin  plastic  membrane  immersed  in  distilled  water 
and  viewed  at  normal  incidence.    For  this  target, 
N(x,y,z)  is  proportional  to  6(x-Xq)  -  6(x-Xq-D) 
where  D  represents  thickness  and  x^  locates  the 
proximal  membrane  surface.    From  eq.  (5)  and  (7), 
the  calculated  power  spectrum,  is  proportional  to 
sin    (2TTfD/c)  so  that  spectral  minima  will  occur 
at  frequencies  separated  by  an  interval  equal  to 
c/2D.    The  observed  spectrum,  shown  in  figure  3, 
exhibits  a  scalloped  shape  in  agreement  with  this 
result.    The  measured  frequency  interval  between 
successive  spectral  minima  is  3.6  MHz  which  cor- 
responds to  a  calculated  thickness  of  360  pm. 
This  value  is  within  5  percent  of  the  measured 
membrane  thickness. 

Spectral  data  were  also  obtained  for  dilute 
solutions  of  plastic  microspheres  with  25  ym 
diameters.    These  dilute  solutions  insure  that 
single  scattering  from  points  is  approximated  over 
the  spectral  measurement  band.    Moreover,  it  is 
reasonable  to  assume  that  their  spatial  distribu- 
tion forms  a  Poisson  process  since  the  random  posi- 
tion of  each  scatterer  should  be  independent  and 
uniformly  distributed  within  the  analyzed  volume. 
Accordingly,  measured  spectra  should  be  described 
by  eq.  (15). 

Figure  4  shows  normalized  spectra  measured  for 
two  microsphere  concentrations.    The  spectral 
curves  agree  very  closely  with  the  u"*  dependence 
expected  from  the  normalized,  non-coherent  spectral 
component  in  eq.  (15).    In  addition,  a  fourfold 
change  in  microsphere  concentration  was  found  to 
produce  a  proportional  6-dB  change  in  spectral 
amplitude.    This  amplitude  change  is  consistent 
with  non-coherent  scattering,  in  which  received 
spectral  power  is  proportional  to  the  mean  number 
of  scattering  particles. 

Clinical  measurements  have  been  carried  out  on 
a  wide  variety  of  normal  and  diseased  structures 
in  the  eye  and  orbit.  In  vivo  data  on  more  than 
200  cases  have  been  acquired  and  digitally  cata- 
logued as  part  of  an  on-going  research  effort.  The 
following  paragraphs  review  some  of  the  observed 
spectral  features  which  relate  directly  to  cases 
treated  in  the  preceding  discussions  of  analytic 
and  experimental  results. 


115 


-20 


-24 


^  -28 


u  -32 

Q 
3 


-36 


-40 


-44 


-48 


-5  2  - 


Af  =  3.6  MH  z 
D  =  360 


J  L 


J  I  I  L 


J  I  L 


5  10 

FREQUENCY  (MHz) 

Fig.  3.    Spectrum  of  echoes  from  plastic  membrane. 


Oi- 


-    -  5 


-10 


-20 


MICROSPHERE  SPECTRA 
25  ^im  DIAMETER 


FULL  CONCENTRATION 
1.7  X  10^/cm 


RAYLEIGH 
RESPONSE  (f^*) 


ONE-QUARTER  CONCENTRATION 
0  43  X  10^ /cm^ 


0.42 
 l_ 


J  L 


Fig.  4. 


7         8        9       10      1 1     12    13    14  15    16  17  18 
FREQUENCY  (MHz) 

Spectra  obtained  for  two  concentrations 
of  dilute  microsphere  suspensions. 


Detached  retinas,  observed  prior  to  the  develop- 
ment of  gross  degenerative  processes,  constitute 
thin  membranes  bounded,  anteriorly,  by  vitreous 
humor  and,  posteriorly,  by  fluid  exudate.  Spectra 
obtained  from  these  structures  display  a  marked 
scalloped  appearance  similar  to  that  seen  for  the 
plastic  membrane  (fig.  3).    This  fact  is  demon- 


strated in  figure  5  which  shows  clinical  spectra 
in  decibels  relative  to  a  glass  ^'late  (dBr)  ob- 
tained from  three  different  cases  of  retinal  de- 
tachments.   Retinal  thickness  can  be  computed  from 
the  frequency  repetition  interval  of  these  spectra. 
Using  a  nominal  propagation  velocity  of  1.5  mm/ps 
for  the  retina,  the  computed  values  in  the  illus- 


116 


Fig.  6.  Clinical  spectra  obtained  for  vitreous  hemorrahages  (9  cases).    Spectral  amplitudes 
have  been  off-set  by  indicated  values. 


117 


trated  cases  are  on  the  order  of  190  ym  and  are 
consistent  with  expected  retinal  dimensions. 

Vitreous  hemorrhages  have  been  studied  exten- 
sively with  the  spectrum  analysis  system.    In  their 
unorganized  state,  these  hemorrhages  contain  ran- 
domly placed  aggregations  of  hemorrhagic  debris 
dispersed  in  sections  of  the  vitreous  humor.  Un- 
organized hemorrhages  consistently  produce  spectra 
whose  amplitudes  increase  markedly  with  increasing 
frequency  as  shown  in  figure  6.    The  rate  of  in- 
crease can  sometimes  be  as  rapid  as  the  f"*  charac- 
teristic associated  with  Rayleigh  scattering: 
this  fact  indicates  scattering  from  a  uniform 
distribution  of  elements  that  are  smaller  than 
35  ym.    Spectral  amplitudes  are  typically  -70  dBr 
near  5  MHz:    these  very  low  levels  can  often  be 
exceeded  by  returns  from  isolated  specular  reflec- 
tors or,  in  the  extreme,  by  the  system  noise  level. 
At  higher  frequencies,  the  spectral  amplitudes 
produced  by  hemorrhagic  debris  are  significantly 
larger  and  exceed  these  background  levels. 

Occasionally,  other  spectral  shapes  have  been 
observed  within  an  examined  vitreous  hemorrhage. 
They  indicate  a  possible  high-density  packing  or 
organization  of  the  hemorrhagic  debris.  The 
presence  of  membranes  has  also  been  found  to  af- 
fect the  observed  spectral  shape.    For  example, 
figure  7  shows  the  scalloped  spectra  received  from 
a  range-gated  volume  which  contained  both  hemor- 
rhagic debris  and  a  thickened  intravitreal 
membrane. 

5.  Conclusion 

The  analysis  outlined  in  the  preceding  sections 
has  proven  extremely  useful  in  designing  the  spec- 
trum analysis  system  and  in  selecting  system  para- 
meters, such  as  gate  dimensions,  for  clinical  ap- 
plication.   It  has  also  formed  a  constructive 
framework  for  interpreting  in  vivo  data,  especial- 


ly for  detached  retinas,  vitreous  hemorrhages,  and 
intra-vitreal  dispersions  of  cholesterol  aggrega- 
tions. 

The  analysis  is  being  extended  to  accommodate 
more  complex  tissue  architectures.    This  effort 
is  proceeding  using  clinically  measured  spectra 
and  subsequent  histologic  preparations  when  avail- 
able.   Of  particular  interest  are  intra-ocular 
tumors  such  as  malignant  melanomas  and  metastatic 
carcinomas. 

Other  topics  under  investigation  include  the 
use  of  spectral  data  to  estimate  tissue  attenua- 
tion characteristics.    Presently,  attenuation 
estimates  are  formed  from  ratios  of  power  spectra 
measured  at  sequential  range  sites  separated  by 
aR.    If  the  examined  tissue  architecture  satisfies 
certain  conditions  (e.g. ,  statistical  homogeneity), 
then  such  ratios  describe  the  attenuation  ex- 
perienced within  AR.    Spectral  data  indicate  that 
this  approach  is  appropriate  for  some  tissues 
(e.g. ,  orbital  fat)  but  is  not  applicable  to  other 
tissues  (e.g. ,  heterogeneous  tumors).    In  these 
latter  cases,  spectral  ratios  might  serve  as  a 
useful  index  of  statistical  homogeneity. 

As  the  clinical  data  base  is  expanded,  these 
investigations  can  proceed  toward  the  ultimate 
goal  of  applying  spectrum  analysis  techniques  to 
supplement  conventional  A-  and  B-scan  ultrasono- 
graphy. 

Acknowledgments 

The  authors  wish  to  acknowledge  the  collabora- 
tion of  D.  Jackson  Coleman  and  Louise  Franzen  in 
the  clinical  aspects  of  this  work.    They  also  wish 
to  thank  Angel  Rosado  for  his  assistance  in  data 
processi  ng. 

Portions  of  this  work  were  supported  by  Public 
Health  Service  Grants  EY-0I212-04  and  EY-01218-04 
from  the  National  Eye  Institute. 


-44  - 


m 

u -52 

Q 

K  -56 


^  -60 
< 


-64 
-68 


^-72 

CO 

-76 


PATIENT  3943 


10 

FREQUENCY  (MHz) 


15 


Fig.  7.  Clinical  spectrum  obtained  for  a  vitreous  hemorrhage  and 
associated  intravitreal  membrane. 


118 


References 

[1]    Waag,  R. ,  Lerner,  R. ,  and  Gramiak,  R. , 

Swept-Frequency  Ultrasonic  Determination  of 
Tissue  Macrostructure ,  in  Ultrasonic  Tissue 
Characterization,  M.  Linzer,  ed.  National 
Bureau  of  Standards  Spec.  Publ .  453,  pp. 
213-230  (U.S.  Government  Printing  Office, 
Washington,  D.C. ,  1976).  ■  •  ■ 

[2]    Lele,  P.  and  Namery,  J.,  A  Computer-Based 
Ultrasonic  System  for  the  Detection  and 
Mapping  of  Myocardial  Infarcts,  in  Proceed- 
ings of  the  San  Diego  Biomedical  Symp.  13, 
121-132  (1974). 

[3]    Sigelmann,  R.  and  Reid,  J.,  Analysis  and 
measurement  of  ultrasound  backscattering 
from  an  ensemble  of  scatterers  excited  by 
sine-wave  bursts,  J.  Acoust.  Soc.  Am.  5^, 
1351-1355  (1973). 

[4]    Lizzi,  F.  L.,  St.  Louis,  L.  ,  and  Coleman, 
D.  J.,  Applications  of  spectral  analysis  in 
medical  ultrasonography.  Ultrasonics  14  (2), 
77-80  (1976). 

[5]    Lizzi,  F. ,  Burt,  W. ,  and  Coleman,  D.  J., 

Effects  of  ocular  structures  on  the  propaga- 
tion of  ultrasound  in  the  eye.  Arch.  Ophthal- 
mol .  84,  635-640  (1970). 


[6]    Lizzi,  F.,  Laviola,  M. ,  and  Coleman,  D.  J., 
Ultrasonic  Tissue  Characterization  Using 
Spectrum  Analysis,  in  Proceedings  of  SPIE/SPSE 
Conference  on  Application  of  Optical  In- 
strumentation in  Medicine  V,  Society  of 
Photo-Optical  Instrumentation  Engineers, 
Palos  Verdes  Estates,  California,  pp.  322- 
328  (1976). 

[7]    Lizzi,  F. ,  Laviola,  M.,  and  Coleman,  D.  J., 
Tissue  Signature  Characterization  Utilizing 
Frequency  Domain  Analysis,  in  Proceedings  of 
IEEE  Ultrasonics  Symposium  1976,  Institute 
of  Electrical  and  Electronics  Engineers,  Inc., 
New  York,  pp.  714-719  (1976). 

[8]    Lizzi,  F.  and  Elbaum,  M. ,  Ultrasonic  Spectrum 
Analysis  Techniques  in  Medical  Ultrasound, 
Riverside  Research  Institute  Technical  Re- 
port, Riverside  Research  Institute,  New  York, 
N.Y.  (in  preparation). 

[9]    O'Neil,  H.,  The  theory  of  focussing  radiators, 
J.  Acoust.  Soc.  Am.  21_,  516-526  (1949). 

[10]    Pratt,  W.    Laser  Communications  Systems, 

pp.  255-262  (John  Wiley  and  Sons,  New  York, 
1969). 


119 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


TISSUE  CHARACTERIZATION  IN  VIVO  BY  DIFFERENTIAL  ATTENUATION  MEASUREMENTS 


Salavatore  Levi  and  Jose  Keuwez 

Ultrasound  Lab  and  F.R.E.S.E.R.H. ,  Hopital  Brugmann 
Universite  Libre  de  Bruxelles,  Brussels,  Belgium 

A  practical  ultrasound  method  has  been  applied  for  the  differentiation  in  vivo 
of  pelvic  tumors.    This  method  has  the  advantage  of  being  applicable  during  usual 
ultrasonic  examination  and  does  not  require  information  about  the  tissues  involved, 
nor  the  surrounding  tissues. 

The  method  is  based  upon  the  fact  that  the  ultrasonic  attenuation  increases  with 
frequency;  the  comparison  of  the  echoes  from  the  posterior  and  anterior  boundaries 
of  the  tumor  leads  to  a  coefficient  of  differential  attenuation.    This  coefficient 
has  been  determined  in  examinations  of  95  gynecological  patients;  the  organs  and 
tumors  explored  were  normal  uteri,  leiomyomas  and  ovarian  cysts.    The  values  of  the 
coefficient  of  differential  attenuation  obtained  for  these  three  categories  of 
tissue  are  sufficiently  separated,  which  enables  us  to  make  a  differentiation. 


Key  words:    Characterization;  differential  attenuation;  tumor;  ultrasound. 


1.  Introduction 

Ultrasound  has  been  shown  to  be  effective  in 
the  diagnosis  of  tumors  and  is  generally  capable 
of  differentiating  between  solid  and  liquid  com- 
ponents.   The  current  acoustic  methods,  which 
have  been  applied  for  a  number  of  years,  give  de- 
pendable results  in  93  percent  of  the  leiomyoma 
cases  and  in  84  percent  of  the  cases  involving 
ovarian  cysts  [1  ]^ . 

Ultrasonic  imaging  has  reached  a  high  level 
of  accuracy,  especially  with  the  application  of 
"grey-scale"  techniques.    However,  other  aspects 
of  diagnostic  ultrasound  have  not  yet  reached 
their  full  potential.    More  precise  information 
on  the  viscoelastic  properties  of  tissues,  espe- 
cially those  of  tumors,  should  be  the  next  step 
in  research  on  diagnostic  methods. 

Quantitative  studies  should  produce  more  mean- 
ingful numerical  data  regarding  the  various  types 
of  tumors  investigated.    Such  data  will  be  rep- 
resentative of  the  acoustical  properties  of  the 
tissues  and  complete  the  subjective  data  produced 
by  imaging. 

One  of  the  most  important  of  these  properties 
is  the  attenuation  of  sound.    The  loss  in  acoustic 
intensity  in  travelling  through  the  biological 
tissues  is  caused  mainly  by  absorption  and  scat- 
tering.   These  causes  may  be  grouped  under  the 
term  "attenuation".    It  is  difficult  to  assess  the 
individual  contribution  of  these  mechanisms  (ab- 
sorption, scattering,  ...)  and  it  is  therefore 
quite  preferable  to  confine  oneself  to  examining 
attenuation  as  a  whole. 

The  importance  of  the  various  mechanisms  is  de- 
pendent on  the  wave  frequency;  therefore  the  at- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


tenuation  is  a  function  of  frequency.    The  attenua- 
tion is  a  characteristic  of  a  tissue,  but  general- 
ly it  is  quite  impossible  to  measure  it  in  vivo 
without  invasive  techniques.    We  chose  the  deter- 
mination of  the  variation  of  the  attenuation  with 
frequency  as  a  basis  for  the  differentiation  of 
tissues  [2-4].    A  similar  concept  has  been  pro- 
posed by  Kossoff  [5,6],  but  no  results  have  been 
published.    For  a  good  understanding,  we  envision- 
ed a  model  similar  to  that  published  by  Hill  [7]. 

2.    Model  and  Analysis 

The  following  analysis  is  approximate  and 
does  not  take  into  account  various  mechanisms 
which  could  be  important.    The  objective  of  this 
analysis  is  to  give  an  idea  of  the  relation  be- 
tween the  ultrasonic  properties  of  tissues  under 
investigation  and  the  coefficient  of  differential 
attenuation  which  will  be  defined  below. 

Generally  any  mass^  consists  of  several  layers 
of  tissue  which  have  their  own  properties;  for 
example,  the  uterine  fibroma  in  which  a  layer  of 
normal  muscle  surrounds  pathological  tissues. 
The  model  (fig.  1)  has  been  constructed  in  order 
to  represent  the  different  layers  of  the  region 
under  investigation  and  of  the  mass.    Each  layer 
is  characterized  by  an  acoustic  absorption  coeffi- 
cient, and  each  interface  by  a  back-scattering 
cross-section  (a),    (o  is  used  here  to  take  account 
of  all  the  mechanisms  which  contribute  to  the  loss 
of  energy  at  this  interface.) 

These  layers  are  separated  into  two  sets;  p 
layers  between  the  transducer  and  the  mass,  the 
mass  being  constituted  of  r  -  p  =  n  layers.  In 
each  layer,  one  of  the  important  mechanisms  in 


2ln  this  paper,  the  term  "mass"  will  include 
the  normal  uterus,  leiomyoma,  and  cyst. 


121 


MASS 

ai 

V  1 

p  layers 


n  layers 


Fig.  1.  Model  of  the  different  layers  of  the 
region  under  investigation. 

which  acoustic  energy  is  degraded  as  heat  by  in- 
ternal friction  of  viscosity  is  the  absorption. 
This  loss  of  acoustic  energy  follows  the  relation 
given  by  the  equation: 


A  =  Aq  exp(-ax) 


(1) 


where  a  is  the  amplitude  absorption  coefficient, 
Aq  and  A  are,  respectively,  the  wave  amplitude  at 
some  reference  point  and  the  amplitude  at  a  fur- 
ther distance,  x,  from  that  reference  point. 

Let  us  consider  a  sound  wave  propagating  from 
the  source  S  to  the  point  R  (fig.  1).    The  ampli- 
tude of  that  wave  at  R  is  given  by 

A(R,v)  =  Ao(v)  o(R,v)  exp(-ai(v)<Si)  (2) 

. . .  exp(-ap(v)6p) 

where  v  is  the  wave  frequency,  a-j  and  s-j  are,  re- 
spectively, the  amplitude  absorption  coefficient 
and  the  thickness  of  the  layer  i;  a{\>)  is  a  factor 
which  describes  the  decrease  of  amplitude  at  the 
several  interfaces.    The  amplitude,  SA,  of  the 
reflected  wave  coming  from  R  on  to  the  transducer 
is  related  as  follows: 


SA(R,v)  =  AqIv)  a(R,v)  exp(-2  ^  cx^^i)  (3) 

i  =  l 

If  (vi,v2)  are  two  different  frequencies,  let  us 
consider  the  ratio  of  the  amplitudes  SA(R,vi)  and 
SA(R,V2);  according  to  eq.  (3),  this  ratio  is  given 
by  the  following  equation: 


SA(P,Vi)          Ao(vi)  o(R,vi) 

In    =  In    +  In   

SA(R,V2)          Ao{v2)  a(R,V2) 

r 

-  2  ^  (ai(vi)-ai(v2)) 
i  =  l 

Similarly,  (see  fig.  1) 

SA(P,Vi)           Ao(vi)  o(P,Vi) 

In    =  In    +  In   

SA(P,V2)             Ao(V2)  ct(P,v2) 


2         (ai (vi )-aT (v2) ) 
i=l 


(4) 


(5) 


Now  we  introduce  J  as  the  difference  of  eqs.  (4) 
and  (5). 

SA(R,vi)  SA(P,Vi) 
J  =  In  In  


SA(P,V2) 
a(R,V2) 


J  =  In  In   ■ 

a(P,Vi)  o(P,V2) 


2    J2     (ai (vi (V2) )6i 
i=p+l 


(6) 


(7) 


Equation  (6)  can  be  written  in  the  following  form: 

(n(vi)  -  n(v2))  -  (a(vi)  -  a(v2)) 


2d 


=  Af)  -  Aa 


where 


a^6]  =  ad    (d  =  thickness  of  the  mass)  (10) 

i=p+l 


In 


t(R,n 


a(P,v] 


2dn(v) 


(IT) 


Examining  eq.  (8),  we  see  that  J/2d  depends  only 
on  the  ultrasonic  properties  of  the  layers  com- 
prised between  P  and  R;  J/2d  is  a  characteristic 
of  the  mass  and  tells  us  about  the  variation  of 
attenuation  with  frequency. 

Let  us  remark  that,  for  the  calculation  of 
the  value  of  J,  we  only  need  the  knowledge  of 
four  amplitudes,  and  no  information  is  required 
about  the  tissues  involved  nor  the  surrounding 
tissues. 

3.    Materials  and  Methods 
We  have  defined  eq.  (6) 


J  =  In 


SA(R,Vi) 
SA(R,V2; 


In 


SA(P,Vi; 
SA(P,V2; 


■  (12) 


which  is  the  basic  equation  for  characterization 
based  upon  differential  attenuation.    For  the  cal- 
culation of  J  it  is  necessary  to  measure  the 
amplitudes  of  the  waves  reflected  by  the  anterior 
and  posterior  boundaries  of  the  mass.    This  mea- 
surement must  be  performed  at  two  frequencies 
(vi  and  V2).    Theoretically,  it  is  possible  to 
undertake  these  measurements  with  a  single  pulse 
[6],  however,  we  have  chosen  to  adapt  eq.  (10)  by 
considering  that  these  amplitudes  are  for  waves 
generated  by  two  relatively  narrow  band  (9.5  MHz) 
transducers . 

These  amplitudes  have  been  measured  with  the 
ultrasonic  equipment  "Kretz  Combison  I"  [2-4]. 
For  each  amplitude  measurement,  all  the  settings 
were  identical  except  that  of  the  master  gain. 
This  last  one  was  adequately  adjusted  to  obtain  an 
amplitude  of  given  level  on  the  A-Scope  for  the 
echo  of  interest  (I).    The  corresponding  reading 


122 


L(I,v)  (in  dB)  was  used  in  the  following  equation 
based  on  eq.  (12) 


Table  1.    Statistical  analysis. 


2d 


L(R,Vi)  -  L(R,V2)  -  (L(P,Vi)  -  L(P,V2))  (13) 


where  d  =  dimension  of  the  mass.    We  call  y  the 
coefficient  of  differential  attenuation.    This  at- 
tenuation coefficient  has  been  determined  in  exami- 
ing  95  gynecological  patients.  The  masses  under  in- 
vestigation were  the  normal  uterus,  leiomyoma  and 
cyst.  The  nominal  frequencies  were  2  MHz  and  4  MHz. 

4.    Results  - 

At  present,  twelve  cases  have  been  labeled  as 
normal  uterus,  and  34  other  cases  have  been  clas- 
sified as  leiomyoma  or  cyst  after  surgery.  We 
plotted  the  values  of  the  differential  attenuation 
coefficient,  y,  versus  the  dimension,  d,  of  the 
mass  (fig.  2).    The  plot  shows  that  y  is  not  in- 
fluenced by  d  and  that  the  y  belonging  to  the 
three  categories  of  tissues  (normal  uterus,  leiomy- 
oma and  cyst)  are  quite  characteristic.  Some 
cysts  have  shown  an  abnormally  high  differential 
attenuation  for  masses  with  liquid  content;  we  also 
obtained  negative  values  for  some  cysts,  however, 
this  peculiarity  has  been  reported  and  explained 
[8].     The  values  of  y  for  the  normal  uterus  are 
higher  than  those  obtained  for  the  leiomyoma,  which 
is  in  agreement  with  the  ultrasonic  properties  of 
these  tissues  and  with  previous  results  [4]. 
The  results  of  the  statistical  analysis  are  shown 
in  table  1. 


y  (dB/cm) 
1-8. 

1.6. 

1.4. 

1.2. 

1.0. 

0.8. 

0.6. 

0.4. 

0.2. 

0.0 

-0.2 


+  normal  uterus 
0  leiomyoma 
X  cyst 


O  OOCP 


d  (cm) 


12 


15 


Fig. 


2.  Plot  of  the  values  of  the  differential 
attenuation  coefficient,  y,  versus  the 
dimension,  d,  of  the  mass. 


Organ 


No. 


Mean 


y(db/cm) 
S.D. 


Normal  uterus 

Leiomyoma 

Cyst 


12 
22 
10 


1.39 
0.57 
0.13 


0.27 
0.22 
0.20 


The  absolute  values  are  not  interesting  by  them- 
selves because  y  is  also  depending  on  the  instru- 
mentation. 

5.    Comments  and  Discussion 

If  in  vitro  it  is  possible  to  imagine  trans- 
mission methods,  it  is  more  difficult  in  vivo 
where  generally  only  reflections  methods  are  ap- 
plicable.   Unfortunately,  it  is  not  enough  to  know 
the  amplitude  of  the  reflected  signals  to  tell 
something  about  the  tissues  involved.    It  is  also 
necessary  to  know  the  properties  and  the  orienta- 
tion of  the  interfaces  which  have  partially  re- 
flected the  sound  waves. 

When  working  in  vitro,  the  samples  can  general- 
ly be  chosen,  cut  and  positioned  as  desired.  In 
vi vo ,  many  constraints  have  to  be  taken  into  ac- 
count.   Theoretically  our  method  avoids  some  of 
them;  for  instance  no  information  concerning  the 
surrounding  tissues  is  required.    However,  diffi- 
culties due  to  the  orientation  of  the  interfaces 
remain.    Sometimes  it  was  difficult  to  obtain  a 
posterior  echo  because  of  the  orientation  of  the 
tumor  with  respect  to  the  abdominal  wall.  The 
orientation  is  also  affected  by  the  cardio- 
respiratory movements.    Besides  the  difficulties 
of  measurement,  one  should  be  aware  that  the  re- 
fraction properties  are  frequency  dependent.  This 
could  cause  important  differences  in  amplitude 
when  using  two  frequencies  vj  and  V2  at  oblique 
incidence.    The  dispersion  of  our  results  can  in 
part  be  explained  by  this  last  discussion  as  well 
as  by  considering  effects  due  to  the  beam  geometry. 

We  characterized  our  transducers  [4];  however 
this  characterization  was  made  in  a  nonattenuating 
medium  (water).    In  our  experiments  the  sound  was 
propagating  into  an  inhomogeneous  and  attenuating 
medium  (for  instance  the  abdominal  wall)  which 
could  strongly  influence  the  beam  geometry. 
Therefore  it  is  illusory  to  pursue  this  discussion 
based  on  our  transducer  characterization.  How- 
ever one  must  be  aware  that  the  beam  geometry  can 
dominate  the  results.    That  is  probably  the  reason 
we  did  not  obtain  coherent  results  when  working 
with  other  transducers.    This  method,  as  it  is  ap- 
plied, considerably  extends  the  duration  of  the 
examination,  which  is  generally  performed  with  a 
full  bladder.    Under  these  circumstances  it  is 
difficult  for  the  patient  to  lie  still.  Techni- 
cal improvements  are  necessary. 

6.  Conclusion 

The  characterization  of  tissues  based  upon  the 
variation  of  attenuation  with  frequency  seems  to 
be  suitable  and  helpful  in  cases  where  scans  alone 
do  not  permit  an  accurate  diagnosis.    However,  we 
would  like  to  emphasize  that  it  was  not  always 
possible  to  determine  the  coefficient  of  differen- 
tial attenuation,  for  several  reasons  developed 


123 


in  the  previous  paragraph.  The  results  reported 
in  this  paper  are  those  obtained  for  frequencies 
of  4  and  2  MHz;  however,  we  also  determined  this 
coefficient  for  frequencies  of  2  and  1  MHz.  The 
results  obtained  in  this  last  case  were  incoherent 
and  the  overlap  between  the  three  categories  of 
tumors  investigated  was  important. 

The  various  inconveniences  described  above 
should  be  resolved  by  a  better  choice  of  trans- 
ducers and  the  way  of  determining  the  various 
amplitudes  required  for  the  computation  of  the 
coefficient  of  differential  attenuation. 

Acknowledgments 

We  wish  to  thank  Professor  R.  C.  Eggleton 
(Indianapolis  Center  for  Advanced  Research, 
Indianapolis,  Indiana)  for  helpful  suggestions 
in  preparing  this  paper,  and  Mrs.  J.  Drake 
for  her  secretarial  assistance. 

References 

[1]    Levi,  S.  and  Delval,  R.,  Value  of  ultrasonic 
diagnosis  of  gynecological  tumors  in  370 
surgical  cases.  Acta  Qbstet  Gynecol  Scand  55 
261-266  (1976). 

[2] 


[3] 


in  Abstracts  of  the  2nd  World  Congress  on 
Ul trasonics  in  Medicine,  p.  81  (Junge  und 
Sohn,  Muchen,  1975). 

[4]    Levi,  S.  and  Keuwez,  J.,  An  Attempt  to  Find 
a  Differential  Attenuation  Coefficient  for 
Ultrasonic  Diagnosis  of  Pelvic  Tumors  in 
Vivo,  in  Ultrasound  in  Medicine,  Vol.  3B, 
p.  1989  (Plenum  Press,  New  York,  1977). 

[5]  Kossoff,  G.,  Display  techniques  in  ultrasound 
pulse  echo  investigations;  a  review,  J.  Clin. 
Ultrasound  2,  61  (1974). 

[6]    Kossoff,  G.,  Reflection  Techniques  for  Meas- 
urements of  Attenuation  and  Velocity,  in 
Ultrasonic  Tissue  Characterization,  M.  Linzer, 
ed.,  National  Bureau  of  Standards  Spec.  Publ . 
453,  pp.  135-139  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

[7]    Hill,  C.  R.,  Echoes  from  Human  Tissues,  in 
Ultrasonics  International  '75  Conference 
Proceedings,  24-26  March  1975,  Imperial 
College,  London,  pp.  20-22  (Guildford  IPC 
Science  and  Technology  Press  Ltd.,  London, 
1975). 


Levi,  S.,  Essai  d'analyse  quantitative  des  [8] 
echogrammes  de  tumeurs  pelviennes;  Journees 
d^Etudes  sur  les  Ultrasons  Appliques  a  la 
Medicine,  Nancy  (1974),  Resume  des  Communi- 
cations,  p.  40. 

Levi,  S.,  An  Attempt  to  Differentiate  Pelvic 
Tumors  in  Vivo  by  Attenuation  Measurements, 


LeCroissette,  D.  H.  and  Heyser,  R.  C, 
Attenuation  and  Velocity  Measurements  in 
Tissue  Using  Time  Delay  Spectrometry,  in 
Ultrasonic  Tissue  Characterization,  M.  Linzer, 
ed.,  National  Bureau  of  Standards  Spec.  Publ. 
453,  pp.  81-95  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 


124 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C. ,  1979). 


STATISTICAL  ESTIMATION  OF  THE  ACOUSTIC  ATTENUATION  COEFFICIENT  SLOPE 
FOR  LIVER  TISSUE  FROM  REFLECTED  ULTRASONIC  SIGNALS 


Roman  Kuc,^ Mischa  Schwartz, ^  Nathaniel  Finby,^  and  Frank  Dain^ 

^Department  of  Electrical  Engineering  and  Computer  Science 
Columbia  University 
New  York,  New  York    10027,  U.S.A. 

^Department  of  Radiology 
St.  Luke's  Hospital  Center 
New  York,  New  York    10027,  U.S.A. 


The  acoustic  attenuation  coefficient  measured  in  dB/cm  is  known  to  increase  linear- 
ly with  frequency  for  liver  tissue.    The  slope  of  this  linear  function,  denoted  by  b, 
has  been  shown  by  other  investigators  to  be  an  indicator  of  tissue  state.    6  is  usual- 
ly measured  from  the  change  in  the  log  spectrum  experienced  by  an  acoustic  pulse  when 
it  is  transmitted  through  the  tissue.    This  paper  presents  research  in  estimating  6 
from  the  reflected  signals  which  are  currently  used  in  the  clinical  environment  to 
generate  diagnostic  images.    The  reflected  signals  from  internal  tissue  structures 
(in  liver  -  the  vascular  and  biliary  systems)  are  distorted  by  the  irregular  reflec- 
tor shapes.    By  modeling  the  roughness  of  the  typical  acoustic  reflector,  the  dis- 
tribution of  the  received  spectra  can  be  derived  and  the  95  percent  confidence  limits 
for  the  measured  B  can  be  calculated.    The  confidence  limits  are  presented  in  terms 
of  the  tissue  size  and  average  reflector  density.    Experimental  results  using  the 
reflected  signals  from  in  vitro  refrigerated  and  formalin-fixed  liver  sections  are 
presented. 


Key  words:    Computer  processing;  estimation  theory;  liver  attenuation;  spectral 
analysis;  statistical  modeling;  ultrasonic  tissue  characterization. 


1.  Introduction 

As  an  acoustic  pulse  propagates  through  liver 
tissue  it  experiences  a  frequency  dependent  attenu- 
ation.   Pauly  and  Schwan  [l]i  observed  that  the 
acoustic  attenuation  was  a  linear  function  of  fre- 
quency and  postulated  that  this  behavior  was  primar- 
ily the  result  of  macromolecular  relaxation  pro- 
cesses.   Lele  Gt_  aj_.  [2]  confirmed  the  linear  fre- 
quency dependence  of  the  acoustic  attenuation  and 
in  addition  experimentally  demonstrated  that  the 
slope  of  the  linear  dependence  increased  for  necro- 
tized tissue.    Lele's  experiments  were  performed  by 
transmitting  an  acoustic  pulse  of  known  shape 
through  a  thin  liver  section  and  observing  the 
change  in  the  log  power  spectrum  as  the  pulse  propa- 
gated through  the  tissue.    In  this  paper  we  discuss 
our  work  in  estimating  the  slope  of  the  linear  fre- 
quency dependent  attenuation,  which  we  denote  by 
the  coefficient  B,  from  the  reflected  acoustic 
signals,  typical  of  those  currently  used  in  the 
clinical  environment  to  generate  the  acoustic 
images. 

The  reflections  observed  from  liver  tissue  are 
caused  primarily  by  the  vascular  and  biliary  sys- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


tems  that  permeate  the  liver  volume.    Because  the 
typical  reflector  encountered  within  the  liver  is 
neither  plane  nor  normally  incident  to  the  inci- 
dent pulse,  the  log  spectra  of  the  reflected  sig- 
nals are  distorted.    In  this  paper  we  account  for 
the  distortion  caused  by  the  irregularly  shaped 
reflector  by  modeling  the  reflector  as  a  random 
linear  filter  having  an  impulse  response  which  is 
a  sample  function  from  a  zero  mean,  white  Gaussian 
process.    From  this  assumed  model  we  can  derive 
the  distribution  of  the  observed  sample  spectrum. 
Using  this  distribution  we  can  calculate  confidence 
intervals  for  the  resulting  estimates  of  e.  Final- 
ly, we  predict  the  sensitivity  of  the  estimator  on 
the  tissue  thickness,  L,  by  assuming  that  the  liver 
tissue  has  a  reflector  density  y  reflectors  per  cm. 
Experimental  results  using  in  vitro  refrigerated 
and  fixed  liver  tissue  specimens  will  be  compared 
to  the  analytically  predicted  performance. 

2.    Acoustic  Attenuation  of  Liver  Tissue 

We  first  describe  the  experimental  procedure 
that  we  use  to  measure  g  from  transmitted  pulses. 
This  is  used  to  verify  the  reflection  measurements. 
The  power  spectrum  of  the  incident  pulse,  denoted 
by  Sx(f)  in  figure  1,  is  found  by  reflecting  the 
pulse  from  a  plane,  normally  incident  reflector 


125 


LIVER 


d 


Power  Spectrum 
dB 


The  average  values  of  6  that  we  obtained  for  in 
vitro  refrigerated  and  fixed  liver  specimens  are 
shown  in  figure  1.    These  values  agree  with  those 
found  by  other  researchers  [1,2].    The  transmis- 
sion technique  works  well  when  thin,  parallel  cut 
sections  of  tissue  are  used.    When  large  hetero- 
geneous tissue  specimens,  or  tissue  specimens 
which  have  irregular  shapes  are  used,  the  propa- 
gating pulse  becomes  distorted  producing  artifacts 
in  the  measurements  [3,4]. 

We  will  extend  this  technique  of  measuring  6  to 
signals  reflected  from  tissue  structures  by  sub- 
tracting the  log  spectra  of  signals  from  two  re- 
flectors within  the  tissue.    But  because  of  the 
distortion  to  the  spectra  caused  by  the  irregular 
reflectors  the  observed  spectral  difference  will 
be  a  distorted  linear  function.    The  coefficient  6 
will  be  estimated  by  fitting  a  straight  line  to  the 
spectral  difference  and  dividing  by  the  round-trip 
distance  between  the  reflector  locations.    In  order 
to  find  a  suitable  estimation  technique  and  the 
confidence  intervals  for  the  resulting  estimate  we 
will  derive  the  distribution  of  the  distortion  to 
the  log  spectrum  in  the  next  section. 

3.    Reflector  Model 


FIXED  LIVER 

(e  =  1.20  dB/cm-MHz) 
REFRIGERATED  LIVER 
(e  =  0.63  dB/cm-MHz) 


f  (MHz) 


Fig.  1.  Transmission  experimental  configuration 
results.    An  acoustic  pulse  propagates 
through  liver  sample  having  thickness  d, 
is  reflected  by  plane,  normally  incident 
reflector  and  propagates  through  sample 
a  second  time.    The  liver  acts  like  a  low 
pass  filter,  attenuating  the  high  fre- 
quencies in  the  power  spectrum.    The  dif- 
ference between  the  power  spectra  is  found 
to  be  linear  with  a  slope  which  is  an  in- 
dicator of  tissue  state.    The  parameter  6 
is  the  observed  slope  divided  of  twice  the 
tissue  thickness. 

♦ 

with  no  tissue  in  the  acoustic  path.    The  power 
spectrum  of  the  pulse  transmitted  through  the  tis- 
sue of  thickness  d  back  to  the  transducer  is  de- 
noted by  SY(f).    The  effect  of  the  tissue  acoustic 
attenuation  can  be  found  by  taking  the  log  spectral 
difference  (in  dB)  of  the  two  spectra.    For  a  line- 
arly frequency  dependent  acoustic  attenuation  ibhis 
will  be  equal  to 


10  logS„(f)  -  logs  (f)  =  2  3df 


(1) 


The  value  of  B,  in  units  of  dB/cm-MHz,  for  a  given 
tissue  can  be  measured  by  dividing  the  observed 
slope  of  the  log  spectral  difference  by  the  acous- 
tic path  length  through  the  tissue.  For  reflected 
signals  the  acoustic  path  length  is  equal  to  twice 
the  tissue  thickness. 


We  propose  to  model  the  irregular  reflector  as 
a  random  linear  filter  with  an  impulse  response, 
denoted  by  r(n)  for  n  =  1,  2,  KTj,  which  is 

sample  function  from  a  zero  mean,  white  Gaussian 
process.    The  probability  density  function  (p)  of 
r(n)  is  then  given  by 


p(r(n))  = 


1 


exp 


r2(n] 


(2) 


For  a  white  Gaussian  process,  the  values  of  r(n) 
and  r(m)  for  sampling  times  n     m  are  independent. 
This  appears  to  be  a  reasonable  assumption  in  this 
work,  considering  the  irregularity  of  the  reflect- 
ing vascular  structures  [5].    This  is  verified  by 
the  experimental  results  to  be  described  later. 

If  Sx(f)  is  the  power  spectrum  of  the  acoustic 
pulse  incident  to  the  reflector  having  the  impulse 
response  r(n),  then  the  power  spectrum  of  the  re- 
flected signal,  SR(f),  can  be  written  as 


SR(f)  =  |R(f)|2Sx(f) 


(3) 


where  |R(f)|2  is  the  power  transfer  function  of  the 
reflector  [6]. 

We  will  estimate  6  from  the  log  spectral  dif- 
ference.   The  logarithm  (in  units  of  dB)  of  eq.  (3) 
is  equal  to 


10  logSj^(f)  =  10  logS^(f)  +  10  log  |R(f)|2 


=  10  logS^(f)  + 


n(f) 


(4) 


(5) 


The  last  equation  indicates  that  the  effect  of 
an  irregular  reflector  is  an  additive  distortion  to 
the  log  spectrum  of  the  incident  pulse  at  the  re- 
flector location.    Furthermore,  since  the  reflector 
impulse  response  is  assumed  to  be  zero  mean,  white 
Gaussian,  it  can  be  shown  [6]  that  the  random 
variables 


iR{fi)r 


(6) 


126 


are  mutually  independent  and  have  an  exponential 
distribution  (the  chi-square  distribution  with  two 
degrees  of  freedom),  at  frequencies  f^  which  are 
separated  by  the  fundamental  frequency,  F  =  l/KIg. 

Since  we  are  seeking  the  distribution  of  the 
distortion  in  the  log  spectral  domain  (in  dB)  we 
set 


n.  =  10  logC.  =  10  log  |R(f.: 


(7) 


Using  the  relation  for  the  pdf  of  a  transformed 
random  variable  [7,8]  the  pdf  of  the  distortion  to 
the  log  spectrum  n-,-  is  found  to  be  equal  to 


p(n  )  =  _  

^       4.34  a^T 


exp 


r  s 


"i 


4.34 


n .  " 
1 

.4.34 


a2T 
r  s 


(8) 


for 


'1 


This  distribution  is  shown  in 


figure  2a. 

If  we  let  h  =  In  a^Ts,  then  a^Tj  =  e^ 
can  then  be  written 


P(n^.) 


4.34 


exp 


n . 

1 

4.34 


-  h 


4.34 


This  last  result  shows  that  the  effect 


Eq.  (8) 


(9) 


the  vari- 


ance of  the  reflector  impulse  response  Oy  on  the 
distribution  of  the  log  spectral  distortion  is  only 
a  shift  by  an  amount  h. 

We  shall  now  describe  the  technique  we  have  de- 
veloped to  partition  the  total  reflected  signal 
from  a  liver  tissue  into  segments  which  are  used 
for  calculating  the  sample  spectra  of  the  returns 
from  the  irregular  reflectors. 

4.    Processing  of  Reflected  Data 

An  acoustic  pulse  having  an  arbitrary  power 
spectrum  Sx(f)  propagates  into  a  tissue  medium 
and  reflections  due  to  changes  in  tissue  acoustic 
impedance  are  detected  by  the  transducer.    A  typi- 
cal reflected  signal  denoted  by  y(n),  is  shown  in 
figure  3.    The  envelope  of  y(n)  is  calculated  by 
rectifying  the  sample  values  and  low  pass  filter- 
ing with  a  digital  nonrecursive  filter.    The  enve- 
lope is  observed  to  contain  a  series  of  N  maxima 
which  have  random  amplitudes  and  occurrence  times. 
We  assume  that  the  occurrence  of  a  maximum,  or 
peak,  in  the  reflected  signal  envelope  indicates 
the  presence  of  an  irregular  reflector.    This  is 
the  same  assumption  used  in  the  equipment  produc- 
ing sonograms  in  the  clinical  environment.  The 
time  series  is  divided  into  N  nonoverl appi ng  data 
segments  corresponding  to  the  signals  returned 
from  the  individual  reflectors. 

The  shape  of  the  reflected  signal  envelope  sug- 
gests a  way  of  segmenting  the  reflected  data.  For 
our  reflected  signal  analysis  the  data  was  segment- 
ed about  the  peaks  in  the  envelope  using  the  loca- 
tions of  the  adjacent  minima  to  delimit  the  seg- 
ment.   Each  segment  is  then  assumed  to  represent 
the  return  from  an  irregular  reflector,  with  the 
position  of  the  envelope  peak  in  the  total  re- 
flected signal  defined  as  the  reflector  location 
within  the  tissue.    This  procedure  produces  a  set 


noi  se 


P(4>i) 


4.34 


I  exp 


'I'i 


exp 


8.68 


N^N(0,ai) 


For  the  emitted 
reflected  signal 


(b) 


Fig.  2.  Reflection  experimental  configuration  and 
procedure.    An  acoustic  pulse  is  emitted 
and  the  reflection  from  the  liver  tissue 
parenchyma  are  detected, 
signal  shown,  a  typical 
contains  a  series  of  peaks  which  have  ran- 
dom amplitudes  and  occurrence  times.  These 
are  most  evident  in  the  envelope.  The 
received  signal  is  divided  into  segments 
delimited  by  the  envelope  mimma.  The 
power  spectra  are  calculated  and  e  is 
estimated  from  the  spectral  differences 
using  least  squares. 

of  nonoverlapping  data  segments  which  are  assumed 
to  be  statistically  independent. 

The  data  segment  corresponding  to  the  k^h  re- 
flector will  be  denoted  by  yk(n),  n  =  1,2, . . . jM^Ts- 
The  sample  power  spectrum,  denoted  by  S|^(f),  is 
calculated  from  y|<(n)  by  using  the  squares  of  the 
cosine  and  sine  transforms  given  by 


T  l/'Vs 
.(f)  =  — 11  E  y(n)cos2TTfn 


k  >  \n=l 


(10) 


X]  y(n)sin  2iTfn 
n=l 


127 


Because  of  the  symmetry  of  the  power  spectrum  it 
suffices  to  consider  the  positive  frequencies  only. 

In  a  manner  similar  to  the  transmission  experi- 
ment we  will  calculate  the  differences  between  the 
sample  spectra  to  estimate  the  value  for  b.    For  N 
segments  there  are  N/2  independent  spectral  dif- 
ferences as  shown  in  figure  3.    From  each  observed 
spectral  difference  we  will  estimate  a  value  for  6, 
denoted  by  6k  foi"  the  kth  spectral  difference. 
The  estimate  for  6  using  the  entire  reflected  sig- 
nal will  be  taken  as  the  average  of  these  individ- 
ual estimates.    A  technique  for  pairing  the  seg- 
ments to  minimize  the  estimator  variance  will  be 


discussed  later. 

The  power  spectrum  of  the  incident  pulse  has 
significant  energy  only  in  a  band  of  frequencies, 
from  fL  (for  lower)  to  fy  (  for  upper).    It  is  in 
this  usable  range  that  the  log  power  spectral  dif- 
ference given  in  eq.  (1)  will  be  approximately 
linear.    Outside  this  range  noise  effects  cause 
random  deviations.    The  set  of  frequencies  f^ 
separated  by  the  fundamental  frequency  spacing 
are  confined  to  be  within  this  usable  range,  i.e.. 


(11) 


LIVER 


Emitted  signal 


Reflected 
signal 


Envelope 


Segments 


Set  of 
log  spectral 
di  fferences 


(b) 


Fig.  3.    Probability  density  functions.    With  the 
rough  boundaries  modeled  as  linear  filters 
with  random  zero-mean  Gaussian  impulse 
responses,  the  values  of  the  power  spec- 
trum at  the  harmonic  frequencies  are 
independent  and  distributed  as  the  func- 
tion shown  in  (a).    The  values  of  the 
power  spectral  difference  at  the  harmonic 
frequencies  are  also  independent  and  dis- 
tributed as  the  function  shown  in  (b). 


In  the  analysis  we  shall  assume  that  there  are  m 
frequencies  within  the  usable  range,  with  fj  de- 
fined as  the  smallest. 

5.    Distribution  of  the  Log 
Spectral  Difference 

Now  let  us  consider  the  log  spectral  difference 
between  the  returns  from  two  irregular  reflectors, 
i  and  j  (a<j),  separated  by  a  distance  d^j.    If  the 
corresponding  power  spectra  of  the  reflected  signal 
are  denoted  by  S|^(f)  and  Sj(f),  then  their  log  spec- 
tral difference  in  dB  at  the  harmonic  frequencies 
is  equal  to 


z.  =  10  logS^(f.)  -  10  logSj.(f.) 
=  10  logs    (f  )  -  10  logSy  (f. 


+  n^(fi)  -  n  (f.; 


(12) 


for  i  =  1 ,  2,  . . . ,  m 

where  ^X^  (f),  k  =  n,  j,  is  the  power  spectrum  of 
the  propagating  pulse  at  the  k^h  reflector.    For  a 
tissue  with  a  linearly  increasing  attenuation  with 
frequency,  the  difference  between  the  propagating 
pulse  log  spectra  is  the  same  as  the  transmission 
experiment  results  given  in  eq.  (1).    In  our  case, 
it  is  equal  to 


10  logS„  (f.) 


for  i  =  1 ,  2 , 


10  logS^  (f^) 


(13) 


If  the  difference  of  the  spectral  distortions  at 
the  harmonic  frequencies  is  defined  as 


^i  =  "£(^) 


n.(f.)  =  n.  „  -  n.  . 


f  (MHz)      then  eq.  (12)  can  be  written  as 


z.  =  23d^jf .  .  s. 


for  i  =  1 ,  2, 


(14) 


(15) 


In  words,  this  last  result  states  that  the  log 
spectral  difference  between  two  reflected  signals 
at  the  harmonic  frequencies  is  a  distorted  linear 
function  of  frequency  with  a  slope  equal  to  the 
value  of  6,  the  coefficient  to  be  estimated,  and 
the  round-trip  distance  between  the  reflectors. 


128 


The  distortion  enters  as  an  additive  noise  term. 
Since  e-j  is  the  difference  of  random  variables 
which  are  independent  and  identically  distributed 
(iid)  at  the  harmonic  frequencies,  ei  is  also  iid 
and  the  variances  are  equal  at  the  harmonic  fre- 
quencies.   Therefore,  we  can2drop  the  i  subscript 
a^d  denote  the  variance  by  a^.    We  can  show  that 
oe  is  independent  of  the  individual  reflector  vari- 
ances and  is  equal  to  62  (dB)^  [7].    Therefore,  the 
distortions  to  the  straight  line  calculated  by 
using  the  log  spectra  from  any  two  reflectors  will 
have  the  same  variance  as  the  reflectors  are  spaced 
farther  apart  (larger  d^j)  the  slope  of  the  log 
spectral  difference  will  increase.    But  since  the 
statistical  roughness  is  assumed  to  be  constant  for 
all  reflectors,  the  distortions  should  remain  the 
same.    We  shall  use  this  fact  later  in  how  we  pair 
segments  to  calculate  the  spectral  difference. 

Since  the  attenuation  coefficient  g  affects  only 
the  slope  in  eq.  (15),  the  mean  value  of  the  dis- 
tortion e-j  can  be  eliminated  by  translating  the 
axes.    This  can  be  ac^compl  i shed  by  destining  new 
variables  Z-j  =  z-j  -  z  and  F.j  =  f-j  -  f  where 


and 


i  =  l 


(16a) 


(16b) 


The  statistical  model  with  zero  mean  distortion 
and  centered  frequency  axis  can  be  written  as 


N'  differences  from  which  to  estimate  6,  where 
N'  2  N/2.    The  statistical  model  using  the  entire 
reflected  signal  can  be  written 


Z.,  =  23d,  F,  + 


^•k 


(20) 


for  i 

k 


1,  2, 
1,  2, 


. ,  m 
. ,  N'  < 


where  Zj|^  is  the  observed  value  of  the  k^h  log 
spectral  difference  (after  the  mean  value  is  sub- 
tracted) at  frequency  F-j  using  signals  from  re- 
flectors spaced  d)^  apart. 

6.    Derivation  of  Maximum  Likelihood 
Estimator  for  6 

The  results  of  the  previous  section  indicated 
that  the  additive  distortions  to  the  log  spectral 
difference  are  closely  approximated  by  independent 
and  identically  distributed  Gaussian  random  vari- 
ables with  mean  zero  and  variance  a^. 

Under  these  conditions  the  maximum  likelihood 
estimate  of  g,  denoted  by  BmL'  is  found  to  be 
equal  to 


N' 


k=l 


k^k 


^ML 


N' 
k=l 


(21) 


where 


for  i  =  1 ,  2, 


m. 


(17) 


The  pdf  of  the  additive  noise  term  with  zero 
mean  can  be  shown  [7]  to  be  equal  to 


1 


1 


4.34 


*i 


+  e 


(18) 


1    i=l  ^ 


(22) 


2d 


i  =  l 

is  the  maximum  likelihood  estimate  of  g  using  only 
the  information  in  the  kth  spectral  difference. 
It  can  be  shown  that  this  estimator  is  efficient, 
i.e.,  it  has  the  smallest  variance  of  any  unbiased 
estimator  [7].    The  variance  of  the  resulting 
estimator  can  be  shown  to  be  equal  to 


We  plot  p(4>-j)  in  figure  2b.    For  comparison  we 
also  s^ow  the  Gaussian  pdf  having  mean  0  and  vari- 
ance Oj.  =  62,  in  dashed  lines.    Because  of  their 
similarity,  we  will  approximate  ifi-j  by  a  Gaussian 
random  variable  having  mean  0  and  variance  a^,  i.e.. 
i.e.. 


Var 


hi]  = 


k=l 


m 

z 

i  =  l 


(23) 


For  notational  convenience  let  us  define  S  as 


1  ^"^l 
P(<f'-i)  =  —     e  ^ 

^     V2tt  a 

e 


(19) 


S  = 


i  =  l 


(24) 


The  statistical  model  given  by  eq.  (17)  de- 
scribes the  log  spectral  difference  between  2  re- 
flectors separated  by  distance  di^.    The  entire 
reflected  signal  contains  H  segments  from  which  we 
can  calculate  N/2  independent  log  spectral  differ- 
ences.   We  will  show  later  that  depending  on  how 
the  segments  are  paired  it  may  be  desirable  to  use 


The  estimator  variance  given  in  eq. 
be  written  as 


(21)  can  then 


Var 


hi]  - 


N' 

4r  d2 
k=l  ' 


(25) 


129 


We  want  to  minimize  the  estimator  variance  by 
properly  selecting  the  segments  used  in  calculating 
the  spectral  difference.    If  adjacent  peaks  are 
chosen,  all  the  du's  will  be  small  and  the  variance 
will  be  large.    Physically,  this  occurs  because  the 
acoustic  attenuation  doesn't  affect  the  signal 
significantly  for  small  tissue  distances  while  the 
irregular  boundary  effects  are  large.    If  we  take 
segments  far  apart  so  that  the  effect  of  acoustic 
attenuation  is  more  apparent,  the  number  of  spec- 
tral differences  N'  becomes  smaller.    We  will  find 
the  optimum  separation  by  considering  a  simple 
tissue  model . 

For  heuristic  purposes  let  us  consider  a  statis- 
tically homogeneous  tissue  L  cm  long  and  having  a 
reflector  density  equal  to  y  reflectors  per  cm. 
The  average  spacing  between  reflectors  is  1/p  and 
the  average  number  of  peaks  in  the  reflected  signal 
is  N  =  yL.    If  the  segments  to  be  used  in  calculat- 
ing the  log  spectral  difference  are  chosen  such 
that  the  distances  between  them  are  constant  and 
equal  to  d^  =  n/y,  N/2  5  n  f  N-1,  the  number  of  in- 
dependent differences  that  can  be  formed  is  equal 
to  N'  =  N-n.    The  minimum  variance  can  be  shown  to 
occur  for  n  =  2N/3.    Then  from  eq.  (25)  we  find  the 
minimum  variance  to  be 


Var 


16 

27  ^ 


(26) 


The  ED  denotes  equal  d^istance  separations  between 
reflectors.    To  relate  this  to  tissue  size  we  sub- 
stitute N  =  yL  to  get 


VarFLj  . 
L  EDJmin 


27 


(27) 


Recalling  that  yL  =  N  we  can  write  eq.  (27)  as 


Varje^J  .  = 
L  EDJmin 


—  N.|  2 

27  ^ 


(28) 


This  shows  that  in  addition  to  decreasing  as  1/N, 
the  variance  also  decreases  as  l/L^.    This  demon- 
strates quantitatively  the  intuitive  result  that 
for  larger  tissues,  the  acoustic  attenuation  be- 
comes increasingly  discernible  relative  to  the 
boundary  distortion.  ^ 

7.    Confidence  Intervals  for  the  Estimator 

In  the  previous  section  we  derived  the  vari- 
ance of  the  ED  estimator  as  a  function  of  tissue 
size  and  reflector  density.    For  Gaussian  random 
variables,  the  95  percent  confidence  interval  is 
defined  as  an  interval  1.96  times  the  square  root 
of  the  variance  about  the  estimate  of  the  param- 
eter.    From  eq.  (27)  the  confidence  interval 
width  is  given  by 


\  „1/2l3/2 


(29) 


with  Cj  =  /27/1 6  S.    For  a  given  reflector  density 


y  the  confidence  intervals  will  indicate  the  tis- 
sue size  required  for  a  desired  estimator  resolu- 
tion. 

We  will  now  present  some  experimental  results 
using  refrigerated  and  formalin-fixed  liver  tis- 
sue specimens. 

8.    Experimental  Results 

A  2.0  MHz  transducer  was  used  to  obtain  data 
from  in  vitro  refrigerated  and  fixed  liver  speci- 
mens.   The  thickness  of  the  refrigerated  liver  was 
6.7  cm.    After  1  month  of  formalin-fixation  the 
tissue  size  was  7.5  cm.    From  transmission  tests 
the  average  values  of  6  were  observed  to  be  equal 
to  0.63  dB/cm-MHz  for  refrigerated  liver  and  1.20 
dB/cm-MHz  for  formalin-fixed  liver.    These  values 
agree  favorably  with  those  observed  for  refriger- 
ated liver  by  Pauly  and  Schwan  [1]  and  for  formalin- 
fixed  liver  by  Lele  [2].    The  usable  bandwidth  was 
determined  from  the  transmission  experiments  to  be 
1.2  to  2.4  MHz.    The  envelope  of  the  reflected 
signal  was  calculated  by  using  a  15  point  lowpass 
filter  with  a  Gaussian  shape  truncated  at  ±  3  a. 
This  particular  filter  was  chosen  because  its  im- 
pulse response  matched  the  envelope  of  the  incident 
pulse.    With  this  filter  the  average  density  of 
peaks,  y,  in  the  reflected  signal  envelope  was 
observed  to  be  equal  to  7.    The  data  were  parti- 
tioned into  non-overlapping  segments  delimited  by 
the  locations  of  the  envelope  minima.    The  same 
power  spectrum  was  calculated  for  each  segment  by 
using  eq.  (10).    The  coefficient  s  was  estimated 
from  the  difference  of  spectra  separated  by  a  dis- 
tance approximately  equal  to  2L/3,  where  L  is  the 
total  size  of  the  tissue. 

In  the  above  analysis  we  were  concerned  with 
the  estimator  performance  as  a  function  of  tissue 
size.    In  order  to  verify  the  predicted  perfor- 
mance of  the  confidence  intervals  with  respect  to 
tissue  size,  small  tissue  sizes  were  simulated  by 
taking  the  equivalent  amount  of  data  from  the  total 
signal  reflected  from  the  liver,  starting  at  the 
beginning  of  the  record.    The  results  then  indi- 
cate the  experimental  performance  of  the  estimator 
as  the  tissue  size  is  increased.    The  confidence 
intervals,  instead  of  being  taken  about  the  esti- 
mated value,  for  purposes  of  comparing  the  results 
of  different  runs,  are  taken  about  the  average 
values  of  B  observed  from  the  transmission  ex- 
periments. 

With  the  15  point  filter  the  average  segment 
size  was  18  samples  at  Tj  =  0.1  ys,  producing  a 
fundamental  frequency  equal  to  0.6  MHz.    Two  in- 
dependent values  per  spectrum,  located  at  1.5  (Fi  = 
-.3)  and  2.1  (F2  =  +.3)  MHz  were  used.    From  eq. 
(24)  the  resulting  value  for  S  is  345.    For  y  =  7 
the  variance  given  by  eq.  (Z7)  is  equal  to 


Var 


K]  - 


(30) 


From  eq.  (29)  the  resulting  95  percent  confidence 
intervals  are  given  by 


CI 


18 


18 


(31) 


The  ED  estimates  as  a  function  of  tissue  size 
for  4  independent  (nonoverlapping)  runs,  denoted 


130 


by  Rl  through  R4,  through  refrigerated  liver  are 
shown  in  figure  4.    The  95  percent  conficence  in- 
tervals are  indicated.    From  the  figure  it  is  noted 
that  the  confidence  intervals  are  a  reasonable  pre- 
diction of  the  estimator  performance:    wider  vari- 
ations and  extreme  values  are  found  at  the  smaller 
tissue  sizes  with  the  variations  decreasing  as  the 
tissue  size  increases.    The  confidence  intervals 
are  rather  wide  indicating  that  large  tissue  sizes 
are  required  to  determine  e  accurately  with  this 
technique. 

The  estimates  for  6  independent  runs,  Fl  to  F6, 
through  fixed  liver  are  shown  in  figure  5.  The 
confidence  intervals  are  centered  about  the  average 
value  of  6  observed  for  fixed  liver.    Again  the 
confidence  intervals  reasonably  predict  the  observ- 
ed performance. 

The  averages  of  the  independent  runs  for  refrig- 
erated and  fixed  livers  are  shown  in  figure  6.  The 
confidence  intervals  have  been  reduced  to  account 
for  the  averages  of  independent  values  (by  1/2  for 
refrigerated  liver,  by  1//6  for  fixed  liver).  Here 
again  the  confidence  intervals  are  consistent  with 
the  observed  results.    Even  for  the  averaged  esti- 
mates the  reduced  confidence  intervals  are  still 

B  (dB/cm-MHz) 

2.8t  \  ■ 


REFRIGERATED  LIVER 
ESTIMATES 


e  (dB/cm-MHz) 


FIXED  LIVER  ESTIMATES 


-1.2 


Fig.  4. 


B  estimates  for  refrigerated  liver.  The 
theoretically  derived  95  percent  confidence 
intervals  are  shown  in  dashed  lines  as  a 
function  of  liver  size.    The  curves  are 
centered  about  the  value  observed  from 
transmission  results.    The  larger  the  liver 
the  more  data  is  produced  and  the  better 
the  resulting  estimate.    Experimental  re- 
sults from  four  runs  Rl  to  R4  are  shown. 
Smaller  liver  sizes  were  simulated  by  tak- 
ing equivalent  sections  of  data  from  the 
total  reflected  signal. 


estimates  for  formalin-fixed  liver.  The 
confidence  intervals  are  shown  in  dashed 
lines  as  a  function  of  liver  size.  The 
curves  are  centered  about  the  value  ob- 
served from  transmission  results.  Experi- 
mental results  are  shown  for  six  runs 
Fl  to  F6. 

overlapped  at  8  cm.    Even  though  the  experimental 
results  shown  here  differ  noticeably  for  refrig- 
erated and  fixed  liver,  the  fixed  liver  estimates 
are  close  to  the  upper  confidence  interval  predict- 
ed for  refrigerated  liver. 

9.  Conclusion 

The  first  order  model  of  the  irregular  reflector 
presented  in  this  paper  offers  a  reasonable  de- 
scription of  the  actual  results  obtained  from  re- 
frigerated and  fixed  liver  samples.    The  resulting 
confidence  intervals,  however,  are  wide  and  overlap 
significantly  indicating  that  a  large  amount  of 
data  is  required  for  a  reasonable  estimator  resolu- 
tion.   The  spread  in  the  estimates  predicted  by  the 
confidence  intervals  are  due  to  the  wide  variations 
possible  in  the  reflected  signal  power  spectrum 
caused  by  the  white  Gaussian  impulse  response  model 
of  the  irregular  reflector.    To  decrease  the  con- 
fidence intervals,  and  hopefully  the  resulting  ex- 
perimental estimates,  a  more  accurate  model  of  the 
typical  reflector  is  required.    In  the  reflector 
model  assumed  here  we  have  ignored  possible  de- 
terministic reflections  and  have  assumed  the  re- 
flector effects  to  be  completely  random.    We  plan 
to  study  an  improved  model,  combining  both  de- 
terministic and  random  effects,  for  which  better 
estimation  techniques  can  be  derived  to  produce 
more  accurate  estimates  from  smaller  tissue  samples. 


131 


(dB/cm-MHz) 


2.4 


2.0 


1.6 


1.2 


0.4 


\  \ 
\  \ 

Averaged  Estimates 

>v 

\  \ 
\\ 

V            0  Refrigerated 
\  liver 

\^     *  Formal  in-fixed 
tv.      \^  liver 

\\  ^^~^^^95%-F 

FORMAl  IN-FIXED 

\,^%-R" 

_REFRIGERA1ED  

 1 — 1  1  1  1  1 — 

1         2  ^3 

- 

^ — f—{  \ — 1^  \     1  \ — 1 

/4      /  5         6        7  (cm 
/  / 
/  / 

// 
1/ 

// 

// 

-1.2 


Fig.  6.    Comparison  averaged  B  estimates  for  refrig- 
erated and  fixed  livers.    The  respective 
confidence  intervals  are  indicated  about 
the  values  observed  in  transmission  experi- 
ments.   The  averages  of  the  4  refrigerated 
liver  runs  and  6  formalin-fixed  liver  runs 
indicate  a  separation  in  their  values. 

Acknowledgment 

This  research  was  made  possible  by  the  loan  of 
the  fast  analog-to-digital  converter  and  paper  tape 
unit  by  the  Picker  Corporation. 


References 

[1]  Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of 
absorption  of  ultrasound  in  liver  tissue, 
J.  Acoust.  Soc.  Am.  50,  792-99  (1971). 

[2]    Lele,  P.  D.,  Mansfield,  A.  B.,  Murphy,  A.  I., 
Namery,  J.,  and  Senepati ,  N. ,  Tissue  Charac- 
terization by  Ultrasonic  Frequency-dependent 
Attenuation  and  Scattering,  in  Ultrasonic 
Tissue  Characterization,  M.  Linzer,  ed., 
pp.  167-96,  National  Bureau  of  Standards  Spec. 
Publ.  453  (U.S.  Government  Printing  Office, 
Washington,  D.C.,  1976). 

[3]    Miller,  J.  G. ,  Yuhas,  D.  E.,  Mimbs,  J.  W., 
Dierker,  S.  B.,  Busse,  L.  J.,  Laterra,  J.  J., 
Weiss,  A.  N.,  and  Sobel ,  B.  E. ,  Ultrasonic 
Tissue  Characterization,  1976  IEEE  Ultrasonics 
Symposium  Proc. ,  76  CH  1120-5SU,  33-43  (1976). 

[4]    Kuc,  R. ,  Schwartz,  M. ,  and  Von  Micsky,  L., 
Parametric  Estimation  of  the  Acoustic  At- 
tenuation Coefficient  Slope  for  Soft  Tissue, 
1976  IEEE  Ultrasonics  Symposium  Proc. ,  76  CH 
1120-5SU,  44-47  (1976). 

[5]    Elias,  H.  and  Sherrick,  J.  S.,  Morphology  of 
the  Liver  (Academic  Press,  New  York,  1969). 

[6]    Jenkins,  G.  N.  and  Watts,  D.  G.,  Spectral 
Analysis  and  Its  Applications  (Holden-Day, 
San  Franci  sco,  1968) . 

[7]  Kuc,  R.,  Statistical  Estimation  of  the  Acous- 
tic Attenuation  Slope  for  Liver  Tissue,  Ph.D. 
Dissertation,  Columbia  University  (1977). 

[8]    Schwartz,  M. ,  Information  Transmission,  Modu- 
lation, and  Noise  (McGraw-Hill,  New  York, 
1970). 

[9]    E.  Parzen,  Modern  Probability  Theory  and  Its 
Applications  (Wiley,  New  York,  1967). 


132 


CHAPTER  5 
SCATTERING 


133 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,   Spec.   Publ .    525    (U.S.  Government  Printing  Office,  Washington,   D.C.,  1979) 


AN  ULTRASONIC  TISSUE  SIGNATURE  FOR  THE  LUNG  SURFACE 


Theodore  L.  Rhynei 

Massachusetts  Institute  of  Technology 

and  Massachusetts  General  Hospital 
Cambridge,  Massachusetts    02139,  U.S.A. 


The  normal  lung  surface  is  modeled  as  a  statistically  rough  surface  composed  of 
the  air-containing  alveolar  sacs  at  the  periphery.    The  ultrasonic  reflection  (scatter) 
from  the  rough  surface  is  analyzed  as  a  random  process  and  a  frequency  signature,  which 
depends  upon  alveolar  sac  radii  statistics,  is  predicted  for  the  reflectance.  Pulse- 
echo  ultrasonic  instrumentation  for  the  measurement  of  the  absolute  reflectance  from 
the  lung  surface  is  presented.    Lung  reflectance  data  confirming  the  lung  surface  fre- 
quency signature  is  presented  for  humans.    Applications  of  the  model  are  discussed  to 
the  detection  of  pulmonary  embolism,  atelectasis,  pulmonary  edema,  and  chronic  lung 
disease. 


Keywords:    Detection  of  lung  disease;  frequency  signature;  statistical  scattering; 
tissue  signature;  ultrasonic  scattering;  ultrasound. 


1.  Introduction 

Ultrasonic  characterization  of  tissue  rep- 
resents a  considerable  challenge  to  the  re- 
searcher in  that  he  must  exploit  essentially 
"second  order"  phenomena  while  current  clinical 
techniques  have  already  exploited  the  "first 
order"  phenomena.    The  "second  order"  phenomena 
(dispersion,  reflection,  scattering,  movement, 
etc. )  are  often  measurable  only  after  suitable 
instrumentation  has  been  developed.    The  re- 
searcher is  especially  challenged  to  select  from 
an  infinity  of  possible  instrumentation  approach- 
es and  plausible  phenomena  in  order  to  arrive  at 
a  clinically  useful  diagnostic  technique.  The 
work  reported  here  is  the  development  of  a  tissue 
characterization  for  the  normal  lung  surface  using 
an  analytical  approach.    The  approach  is  quite 
possibly  as  significant  as  the  results  since  it 
offers  a  path  through  the  complex  decisions  in- 
volving instrumentation  design,  study  of  phenomena, 
and  definition  of  the  useful  measurement. 

The  analytical  approach  used  here  is  summarized 
in  the  four  interactive  steps  outlined  below.  The 
interplay  of  each  step  can  be  followed  as  the 
normal  lung  model  is  developed. 

1)  Calibration  of  the  transmitter-receiver 
circuitry  and  transduction  process  to  either  an 
absolute  or  traceable  standard  for  measurement  of 
tissue  acoustic  properties  that  are  independent  of 
the  instrumentation. 

2)  Design  of  the  transmitted  acoustic  signal 
waveform  to  accentuate  the  desired  tissue  phe- 
nomena. 

3)  Development  and  validation  of  analytical 
models  for  the  tissue  target  which  characterize 


^Current  Address:  Bolt,  Beranek  and  Newman,  Inc., 
50  Moulton  Street,  Cambridge,  MA  02138. 


the  measurable  acoustic  phenomena  of  tissue  in  the 
normal  and  diseased  states. 

4)    Development  of  signal  processing  operations 
performed  by  the  instrumentation  to  detect  and 
quantify  echo  parameters  that  characterize  the 
normal  and  diseased  states  of  the  tissue. 

2.    Normal  Lung 

The  lung  is  a  very  large  organ  whose  surface 
lies  within  a  few  centimeters  of  the  skin  over 
most  of  the  chest.    As  shown  in  figure  la,  a  trans- 
ducer coupled  to  the  skin  may  direct  ultrasonic 
waves  through  the  soft  tissues  of  the  intercostal 
space  that  lies  between  adjacent  ribs.    If  a  suffi- 
ciently small  transducer  is  utilized,  the  lung 
surface  can  be  examined  with  no  interference  from 
nearby  ribs. 

A  normal  lung  has  a  gross  density  that  is  ap- 
proximately one  third  that  of  water.    Thus,  the 
lung  is  approximately  two  thirds  air  and  one  third 
soft  tissue.    At  the  periphery  the  schematic  of 
the  magnified  lung  surface  in  figure  lb  indicates 
the  relationship  of  the  air-containing  alveolar 
sacs  to  the  soft  tissues  that  surround  the  sacs. 
The  alveolar  sac  is  the  end-respiratory  airway  and 
is  divided  up  into  the  familiar  alveolar  chambers. 

The  marked  difference  in  acoustic  impedance  be- 
tween the  air  and  the  surrounding  soft  tissue 
causes  nearly  perfect  reflection  at  the  air  to  soft 
tissue  boundary.    For  our  purposes  we  postulate  an 
"ultrasonic"  lung  surface  which  is  a  roughened 
planar  surface  composed  of  the  outer  walls  of  the 
multitude  of  peripheral  alveolar  sacs  that  lie  just 
inside  the  pleural  membrane. 

The  physical  lung  surface  (as  opposed  to  the 
"ultrasonic"  surface)  is  a  pleural  membrane  which 
forms  a  unique  bearing  with  a  second  pleural  mem- 
brane that  is  attached  to  the  inside  of  the  chest 


135 


Lung  Surface  (Pleura) 


Skin  and  Soft  Tissue 
(a  ) 

Pleural  Surface 


Alveolar  Sac 
(Diameter  «275;i.m) 


(b) 

Fig.  1.    Schematic  representation  of  the  chest 
and  lung  surface;  (a)  cross  section  of 
the  chest  wall;  (b)  enlarged  cross  sec- 
tion of  the  lung  surface. 

wall.    During  respiration  the  lung  moves  lateral- 
ly within  the  chest  on  this  pleura-fluid-pleura 
bearing.    Since  the  bearing  is  composed  of  soft 
tissue  and  is  only  a  few  tens  of  a  micrometer 
thick,  it  does  not  contribute  a  significant  echo 
in  comparison  to  the  "ultrasonic"  lung  surface. 
Furthermore,  the  lateral  motion  of  the  lung  per- 
mits the  lung  surface  to  be  scanned  during  respira- 
tion by  a  transducer  held  stationary  with  respect 
to  the  chest  wal  1 . 

The  geometry  of  the  lung  examination  is  provid- 
ed in  figure  2.    A  radiating  and  receiving  trans- 


Transducer  Aperture  Rough  Surface  of 

Infinite  Extent 


Fig.  2.    Geometry  for  the  examination  of  the  lung 
surface  by  a  transducer  with  a  circular 
aperture. 

ducer  located  at  the  skin  is  represented  by  a  circu- 
lar disk  which  is  the  aperture  of  the  transducer. 
The  radiating  and  receiving  disk  observes  the  lung 
surface  through  the  soft  tissue  of  the  intercostal 
space,  which  is  a  uniformly  lossy  media.  The 


"ultrasonic"  lung  surface  is  a  rough  surface  normal 
to  the  disk  and  of  infinite  extent.    The  rough  sur- 
face may  move  laterally  with  no  change  in  range, 
thus  presenting  new  portions  of  its  surface  to  the 
area  directly  underlying  the  disk. 

3.    Calibrated  Instrumentation 
and  Signal  and  Design 

Wave  propagation  between  the  radiating  (and  re- 
ceiving) aperture  and  lung  surface  forms  a  central 
issue  in  the  calibrated  measurement  of  echoes  re- 
flected from  the  lung  surface.    If  we  further 
simplify  the  situation  represented  in  figure  2  to 
a  lossless  media  and  a  flat  perfectly  reflective 
plane,  we  have  a  classic  radiation  (or  diffraction) 
problem.    By  virtue  of  the  finite  size  of  the 
radiating  and  receiving  disk  plus  the  wave  speed 
of  the  media,  waves  launched  by  the  disk  produce 
filtered  echoes  as  they  reflect  from  the  plane  and 
return  to  the  disk.    The  radiation  (or  diffraction) 
filter  response  is  given  in  Appendix  A  as  developed 
in  the  literature  [1]^.    If  the  response  is  graphed 
as  in  figure  3  for  a  representative  situation,  we 
see  that  the  radiation  filter  is  a  high  pass  filter. 
The  nearly  flat  response  of  the  radiation  filter 
depends  on  the  selection  of  aperture  size,  fre- 
quency, and  range  so  that  the  surface  lies  within 
the  near-field  of  the  transducer.    This  is  the 
case  for  the  lung  surface  measurements.  Sub- 
sequently, the  radiation  filter  will  be  used  both 
for  calibration  and  adjustment  of  lung  surface 
data. 


cr  ■ 
-5  -  . 

-6  1— — ^  ^  1  1  1  I       I       I  I 

0      I       23456789  10 
Frequency  MHz 

Fig.  3.    Frequency  response  of  the  radiation 
filter  between  a  radiating  disk  of 
radius  0.0125  inch  and  a  flat  perfectly 
reflecting  plane  1.5  cm  distant. 

The  pulse  echo  instrumentation  is  diagrammed  in 
the  network  of  figure  4.    It  consists  of  a  trans- 
mitter, a  receiver,  and  a  transducer,  each  linear-  '< 
ly  coupled  to  a  common  transmission  system.  The 
transmitting  waveform       excites  the  system.  The 
waveform  V^(t)  propagates  through  the  transmission 
line  to  the  transducer.    The  transducer  couples 
the  electrical  domain  to  the  acoustic  radiation 
field.    Energy  launched  by  the  radiating  disk  of 
figure  2  is  represented  as  energy  consumed  in  the 
radiation  resistance.    Conversely,  echoed  waves 
striking  the  disk  are  represented  by  the  wave  ' 

^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


136 


Radiation 
Impedance 
+ 

Wave 
Potential 


Transducer 
Loss 


Lossless 

Lossless 

Matching 

Transducer 

Network 

I   Transmission  Line 
Transformer  of  Impedance  Rg 


Receiver 


Outputs 


Fig.  4.    Schematic  of  pulse-echo  ultrasonic  instrumentation. 


potential  Vg(t).    The  radiated  waveform  is  Vt(t) 
as  filtered  by  the  transmission  line  and  the 
transducer.    The  received  waveform  Vf-(t)  is  the 
wave  potential  Vg(t)  as  filtered  once  again  by  the 
transducer  and  transmission  line. 

If  the  transmitter  (plus  receiver)  matches  the 
impedance  of  the  transmission  line,  then  the  filtra- 
tion of  the  transmission  line  consists  of  simple 
time  delays  of  the  transmitted  and  received  sig- 
nals.   Also, the  impedance  presented  to  the  trans- 
ducer is  Rq  in  series  with  V-t-(t).    Thus,  matching 
effectively  removes  the  transmission  line  from  the 
network  of  figure  4.    The  net  effect  of  the  in- 
strumentation is  the  filtration  of  transmitted  and 
received  signals  by  the  transducer. 

In  order  to  measure  transducer  response  and  the 
magnitude  of  echoes  from  a  target,  it  becomes  im- 
portant to  define  a  measure  of  absolute  signal 
loss.    This  is  done  by  energy  conservation  of  the 
energy  in  the  transmitted  wave  V^(t),  a  portion  of 
which  ultimately  returns  as  the  received  echo 
Vp(t).    Loss  is  defined  here  as  the  ratio  of  the 
exchangeable  power  of  the  transmitter  to  the  re- 
ceived power  in  eq.  (1).    This  is  more  convenient- 
ly expressed  as  the  gain  (always  less  than  unity) 
or  transfer  gain  in  eq.  (2)  which  can  be  expressed 
in  decibel  or  magnitude  notation. 

(V2)/(4R  )  V. 

Loss  (dB)  =  10  logio —  20  logiQ—  (1) 

(Vp/(R  )  2V 


2V 

gain  =  — ^  ,  (2) 


Since  the  transmitted  acoustic  waveform  is  de- 
termined by  V|.(t),  we  are  free  to  choose  it  so  as 
to  accentuate  some  property  of  the  target.  Here, 
the  magnitude  of  the  lung  surface  echo  is  the  im- 
portant property.    Furthermore,  the  returned  lung 
echo  will  be  corrected  by  the  radiation  filter 
and  transducer  filter  magnitudes.    Therefore,  we 
select  a  transmitted  waveform  which  is  a  carrier 
burst    of  sufficient  duration  to  allow  the  trans- 
ducer and  the  radiation  process  to  reach  their 
steady-state  or  Fourier  responses.    The  trans- 
mitted pulse  waveform  is  completely  described  by 
the  magnitude,  frequency,  and  duration  of  the 
sinusoidal  carrier.    Conversely,  the  carrier 
burst  must  be  sufficiently  short  in  time  to  allow 
range  discrimination  of  the  lung  echo  from  other 


echoes.    The  values  for       and       utilized  in  loss 
calculations  are  the  instantaneous  values  of  the 
envelopes  of  carrier  waves. 

The  transducer  frequency  response  is  experi- 
mentally determined  as  an  absolute  quantity  using 
the  selected  signals,  the  definition  of  loss,  the 
radiation  filter  correction  (fig.  3),  and  an  ex- 
perimental geometry  as  in  figure  2  (with  a  flat 
perfectly  reflective  planar  test  target)  [2].  It 
is  interesting  to  note  that  absolute  dosimetry  may 
be  determined  as  the  power  dissipated  in  the  radia- 
tion resistance  of  figure  4  [2,3].    The  transducer 
of  figure  4  provides  a  transformer  and  a  lossless 
matching  network  in  addition  to  the  transducing 
plate  and  associated  radiation  and  dissipation 
loads.    Modeling  of  the  transducing  disk  is  ex- 
tensively discussed  elsewhere  [2,3]  where  it  is 
shown  that  a  simple  model  for  transducer  gain 
exists  at  half-wave  resonance.    Also,  it  is  shown 
that  optimum  loop  sensitivity  results  from  adjust- 
ment of  the  transformer  for  impedance  matching  of 
transmitter  to  transducer  impedances.    The  adop- 
tion of  a  transmission  system  impedance  plus 
specification  of  transducer  transfer  function 
could  form  the  basis  of  standardization  for  pulse 
echo  instrumentation. 

4.    Scattering  Model  for  the  Lung  Surface 

Returning  to  the  measurement  of  the  reflection 
from  the  lung  surface  depicted  in  figure  2,  we  are 
motivated  by  the  microscopic  structure  to  model 
the  lung  as  a  randomly  rough  surface.    The  essen- 
tial difference  between  the  flat  surface  of  figure 
3  and  a  rough  surface  is  that  the  rough  surface 
has  a  finite  thickness  or  scattering  depth  and  is 
said  to  be  range-spread.    Mathematically,  this  is 
described  in  eq.  (3)  as  an  integration  over  all 
range  of  some  range-spread  function  m(')  with  the 
impulse  response  of  the  radiation  coupling 
f(',-,-)^  which  is  the  time  domain  pair  to  eq. 
(A-1). 

00 

r(t)  =  f  f(r,a,t)  m(r)  dr  (3) 


For  a  "thinly  rough"  surface  we  may  expand  the 
function  f(r,a,t)  about  some  fixed  range  (rp), 
make  a  change  of  variables  for  fixed  wave  speed  c, 
and  thus  turn  the  equation  into  the  time  convolu- 
tion in  eq.  (4). 


137 


r(t) 


J  f(r^,a,t-T) 


c_  m(CT) 
2  2 


dx 


(4) 


Equation  (4)  is  highly  significant  in  that  the  re- 
ceived signal  r(t)  is  the  result  of  a  linear  filtra- 
tion in  cascade  with  the  radiation  filter  of  eq. 
(A-1).    The  frequency  equivalent  of  eq.  (4)  is 
simply  a  product  of  transfer  functions  in  eq.  (5), 
where  G(f)  is  the  transfer  function  associated 
with  m( • ) • 


R(f)  =  F(r,a,f)  G(f) 


(5) 


The  filter  G(f)  and  its  transform  pair  he 
m(%CT)  represent  the  filtration  of  a  randomly 
rough  surface  and,  therefore,  are  themselves  ran- 
dom.   In  a  sense  the  precise  G(f)  depends  upon 
the  particular  ensemble  of  scattering  elements 
that  lie  in  the  field  pattern  of  the  transducing 
disk  of  figure  2.    The  value  of  the  Fourier  func- 
tion G(f)  is  the  magnitude  (and  phase)  of  the  re- 
flection due  to  a  long  carrier  burst.    By  allow- 
ing the  transducer,  radiation  filter,  and  reflec- 
tion filter  to  reach  steady-state  as  discussed 
above,  one  may  make  direct  measurement  of  the 
magnitude  of  G(f).    The  measured  value  will  vary 
as  different  ensembles  of  scattering  elements 
sweep  in  front  of  the  transducer.    The  charac- 
terization of  the  lung  surface  lies  in  relating 
the  statistical  properties  of  the  lung  surface  to 
the  probability  density  function  (and  its  mean  and 
variance)  for  the  random  variable  |G(f)|. 

A  randomly  rough  surface  representing  the  normal 
lung  consists  of  a  planar  array  of  the  alveolar 
sac  wall  segments  shown  in  figure  1(b).  Between 
adjacent  alveolar  sac  chambers  cusp-like  regions 
are  formed  which  contain  soft  tissue  which  readily 
support  wave  propagation.    The  depth  of  these 
cusps  constitutes  the  thickness  or  scattering 
depth  of  the  "ultrasonic"  lung  surface.    The  lung 
surface  is  characterized  as  having  a  mean  scatter- 
ing depth  (wq)  and  standard  deviation  of  scatter- 
ing depth  [o^).    Figure  1(b)  indicates  that  the 
scattering  depth  Wq  is  approximately  equal  to  the 
mean  radius  of  the  peripheral  alveolar  sacs.  If 
the  wavelength  of  the  ultrasound  is  very  much 
greater  than  Wq  one  expects  the  surface  to  reflect 
as  though  it  were  flat.    Conversely,  at  some  criti- 
cal wavelength  related  to  the  mean  scattering 
depth  Wq  one  would  expect  destructive  interference. 
The  degree  of  the  destructive  interference  would 
of  course  depend  upon  the  regularity  of  the  scat- 
tering surface  with  greater  destructive  inter- 
ference occurring  for  smaller  a^. 

Using  the  above  model  for  the  lung  surface  plus 
several  additional  assumptions  the  probability 
density  function  plus  its  mean  and  standard  devia- 
tion for  |G(f)|  are  predicted  in  Appendix  B.  The 
density  function  is  a  Ricean  function  whose  shape 
depends  upon  Wq,  a^,  and  the  frequency  f.  The 
most  significar[t  measurable  properties  of  |G(f)| 
are  the  mean  (Z)  and  standard  deviation  (a^)  given 
in  figure  5  as  functions  of  frequency.    At  lower 
frequencies  the  curves  approach  those  of  a  perfect 
reflector  (OdB).    Conversely,  there  is  a  destruc- 
tive interference  minimum  at  a  frequency  where 
one-half  a  wavelength  equals  the  mean  scattering 
depth  Wq.    Figure  5(a)  demonstrates  the  dependence 
of  the  dip  frequency  on  Wq.    In  figure  5(b)  we 
note  that  the  depth  of  the  dip  increases  for  great- 
er regularity  in  the  scattering  depth  (smaller  c^) . 


?-5 


■^-10 

o 

tr 

1-20 
o 

^-25 
qI 


-30 


1  1  1  1  1  \  1  1  r 


Mean  Z 


leO^nn 


Standard  Deviation  o-. 


J_ 


4  5 
Frequency  MHz 

(a) 


4  5 
Frequency  MHz 
(b) 


Fig.  5.    Families  of  curves  for  the  mean  Z  and 
standard  deviation       of  the  predicted 
echo  amplitude,  dB  versus  frequency,  (a) 
curves  for  two  values  of  scattering 
depth  Wq  and  fixed  a^;  (b)  curves  for 
two  values  of  a^^  and  fixed  Wq. 

5.    Experimental  Results 

Experimental  measurement  of  the  lung  surface 
reflection  can  be  made  to  confirm  the  above  model. 
Measured  values  are  corrected  for  radiation, 
transducer,  transmitted  voltage,  and  bulk  tissue 
effects  as  in  eq.  (6).    The  resulting  experimental 
value  (Y)  is 

,(l/20)2rf0.897  IQ-^ 


2V  10' 
r 


|F(r,a,f)|  |T(f)P 


(6) 


is  a  measure  of  the  absolute  reflectivity  |G(f)|  of 
the  lung  surface  at  the  frequency  f.    In  the  equa- 
tion Vr  and  Vt  are  envelopes  of  sinusoids  as  dis- 
cussed earlier.    The  function  |T(f)|2  is  the  mea- 
sured transducer  loop  transfer  function  and 
F(r,a,f)  is  the  radiation  transfer  function  for 
range  r  and  frequency  f.    Correction  is  made  for 
bulk  tissue  loss  using  an  attenuation  constant  of 
0.897  dB/cm. 

By  averaging  over  many  values  of  Y  at  a  single 
frequency  for  successive  pulse  echo  transmissions 
with  the  lung  surface  scanning  before  the_trans- 
ducer,  we  arrive  at  the  mean  reflectance  Y  for 


138 


m  -5 
o-  10  - 


=£-15- 


-20 


g-25 
-30 


Mean  Y 


Standard  Deviation  a„ 


4  5  6- 

Frequency  MHz 

Fig.  6.    Experimental  data  for  the  mean  and 

standard  deviation  of  lung  reflection  Y 
for  a  normal  human  volunteer. 

that  frequency.    Similarly,  one  may  compute  the 
standard  deviation  (a^)  for  Y  over  the  data  for 
a  single  frequency,    when  multiple  frequencies 
are  examined  and  the  data  reduced  for  Y  and  oy, 
the  experimental  data  in  figure  6  is  arrived  at. 
Note  that  the  destructive  dip  is  present  near 
5.2  MHz  and  the  curve  rises  toward  0  dB  at  lower 
frequencies.    A  more  extensive  discussion  shows 
close  correlation  between  the  predicted  proba- 
bility density  function  of  Appendix  B  and  ex- 
perimental histograms  [2,31.    These  data  confirm 
the  analytical  model  of  tissue  thus  completing 
step  3  in  the  approach  to  tissue  characterization. 

6.  Applications 

The  final  element  in  the  approach  to  tissue 
characterization  is  to  define  the  signal  process- 
ing functions  of  figure  4  to  perform  some  clini- 
cally useful  application.    Two  areas  of  applica- 
tion are  suggested  for  the  normal  lung  tissue  sig- 
nature developed  here,  they  are:    a)  detection  of 


acute  infiltrative  disease  and  b)  detection  of 
chronic  pulmonary  disease. 

The  normal  lung  surface  strongly  reflects  the 
ultrasound  due  primarily  to  the  presence  of  air 
at  the  periphery.    Several  diseases,  including 
pneumonia,  edema,  atelectasis,  and  pulmonary  em- 
bolism, which  acutely  alter  the  air  content  at  the 
periphery  should  be  detectable  as  a  departure  from 
the  normal  model.    Preliminary  clinical  evidence 
suggests  that  reductions  in  lung  reflectivity  from 
10  dB  to  30  dB  are  associated  with  these  acute  dis- 
eases [2,3].    An  optimal  method  of  detecting  the 
reduction  in  reflection  has  been  suggested  [2,3] 
utilizing  the  predicted  probability  density  func- 
tion to  determine  a  sufficient  statistic  which  is 
the  root  mean  square  reflection.    An  ultrasonic 
instrument  embodying  this  technique  is  shown  in 
figure  7.    The  performance  of  this  detection 
scheme  can  be  evaluated  in  terms  of  probability 
for  false  positive  and  false  negative  with  the 
testing  threshold  (T)  set  to  optimize  performance. 

The  chronically  abnormal  lung  can  be  postulated 
to  produce  reflection  functions  differing  from 
those  of  figures  5  and  6  due  to  rearrangement  of 
tissue  structures  at  the  periphery  of  the  lung. 
Diseases  such  as  emphysema  and  pleural  diseases 
are  of  clinical  intere^st.    Figure  8  demonstrates 
the  reflection  data  (Y,ay)  for  a  subject  with 
severe  chronic  pulmonary  disease.    The  destructive 
interference  dip  has  been  suppressed  presumably  by 
the  higher  variance  in  the  scattering  depth  (a^) 
of  the  diseased  structures  at  the  periphery.  In- 
deed, if  a  greater  value  of  a^j  is  used  in  the  model 
of  figure  5(b)  the  dip  becomes  obliterated.    A  sug- 
gested application  to  chronic  disease  is  to  fit 
predicted  curves  to  experimental  data  and  take  the 
resulting  scattering  parameters  Wq  and       as  in- 
dices of  the  lung  surface  [2,3]. 

The  above  applications  depend  upon  detecting 
departures  from  the  normal  model.    Further  develop- 
ment of  this  tissue  characterization  approach  sug- 
gests developing  detailed  models  (step  3)  for  the 
diseased  states  so  that  further  clinical  applica- 
tions can  be  defined  (step  4). 


Transducer 

Lung  Surface 

Transnnitter 


•lF(r,a,f)l  Vt  lT(f)|2- 

.7; 


Meter 


Light  if 
Less  Than 
Threshold 


RMS  Envelope 


Comparator 


2  ll/2 


Sample  Lung  Echo 


T  Threshold 


Fig.  7.    Ultrasonic  instrument  for  detecting  acute  lung  disease  using 
a  sufficient  statistic  for  testing  against  a  threshold. 


139 


CD 

"°-5 


;-io 


-15 


^-25  - 


-30 


"I  1  1  1  1  1  r 


Mean  Y 


2^-20-    -_  - 


Standard  Deviation , 
J  I  1  L_ 


J  L 


4  5  6 

Frequency  MHz 


Fig.  8.    Experimental  data  for  the  mean  and 

standard  deviation  of  lung  reflection  Y 
for  a  subject  with  severe  chronic  pulmo- 
nary disease. 


Appendix  A 
Radiation  Filter  Response 

The  radiation  filter  response  reported  in  the 
literature  [1]  and  tabulated  [2]  is  given  in  eq. 
(A-1)  for  the  definitions  in  eqs.  (A-2).    It  is  an 
exact  solution  to  the  radiation  coupling  between 
two  coaxial  disks  spaced  at  distance  2r  apart. 


F(r,a,f)  =  [cos(2Trft3)  +  j  sin(2TTft3 )] 


(A-1 


c^ 

-  t3ti, 

;.2 


"tu    +  t, 


Lt,  +  ti 


jQ(2^ft3)    +  jJi(2TTft3 


rtu  +  t, 


+  —  tf 


In  the  equations  c  is  the  wave  speed,  f  is  the 
frequency,  a  is  the  disl<  radius,  and  r  is  the 
range  between  disl<  and  plane.    The  functions  Jq, 
are  Bessel  functions  of  the  first  l<ind. 


Appendix  B 
Detailed  Scattering  Function 
for  the  Lung  Surface 

This  appendix  is  a  summary  of  a  more  detailed 
development  of  the  scattering  function  for  the 
lung  given  elsewhere  [2,3].    The  range  spread 
function  m(r)  is  considered  to  be  the  incremental 
reflectance  of  the  rough  surface  due  to  surface 
elements  between  r  and  r+dr  in  range.    In  order  to 
simplify  the  development  the  following  assumptions 
are  made: 

1)  Reflection  is  a  first  order  scattering 
process  involving  no  shadowing  of  reflective  ele- 
ments nor  multiple  reflection  among  elements. 

2)  The  dimensions  of  roughness  elements  are 
small  compared  to  the  diameter  of  the  transducer. 

3)  The  soft  tissue  to  air  interfaces  perfect- 
ly reflect  with  no  other  loss  mechanism. 

4)  The  disl<  is  normal  to  the  surface. 

5)  The  scattering  depth  is  thin  compared  to 
variations  in  the  radiation  transfer  function. 

6)  The  incremental  reflectance  from  an  in- 
dividual scattering  element  is  of  constant  ampli- 
tude over  the  depth  of  the  element. 

7)  The  scattering  elements  are  statistically 
independent. 

8)  The  lung  surface  approaches  a  flat  perfect- 
ly reflective  plane  as  the  frequency  approaches 
zero. 

The  transfer  function  for  an  individual  scat- 
tering element  of  depth  Wjv,  is  given  in  eq.  (B-1). 
The  overall  transfer  function  is  a  sum  over  a 
large  number  of  individual  scattering  elements  in 
eq.  (B-2). 


L(f  ,wni) 


j^TTf 


-j2TTf2wni 


1  -  e 


(B-1 


1  .   +   

tu    +    tl  t.    +  t; 


J^(27Tft3; 


Jl  (27Tftc 


+  higher  terms 

tl  =  2r/c 

[432  +  4r2 


27Tft. 


+  jJl(2TTft: 


(A-2a) 
(A-2b) 


F(f)  =  h  E  L(f,wm) 
M=l 


(B-2) 


Invoking  the  central  limit  theorem  G(f)  becomes  a 
Gaussian  random  variable  with  independent  real  and 
imaginary  parts  with  means  R^,  Qq  and  variance  a^. 
When  the  value  of  k  in  eq.  (B-1)  is  adjusted  to 
normalize  the  magnitude  as  in  assumption  8,  the 
essential  values  for  Rp,  Qq  and       are  arrived  at 
in  eqs.  (B-3)  and  (8-4). 


4w2(2Trf)2 


1  +  e 


-4aw2(2Trf  )2 

r2 


t3   =  h{t2   -  tl) 
=  h[t2   +  tl) 


(A-2c) 
(A-2d) 


.  .^  -2aw2(2Trf)2 

(2wo2uf)   ^2—^ 

2  cos   e 


(B-3) 


140 


8w2(2TTf)2 


-4aw2(2Trf )- 


If  we  are  concerned  only  with  the  magnitude  Z 
of  G(f)  the  probability  density  function  becomes 
Ricean  as  in  eq.  (B-5).  The  mean  and  a  variance 
of  the  magnitude  are  given  in  eqs.  (B-6)  and  (B-7) 
respectively.  The  functions  lo(')  and  lii-)  are 
modified  Bessel  functions. 


References 

^"4)      [1]    Rhyne,  T.  L.,  Radiation  coupling  of  a  disk 

to  a  plane  and  back  or  a  disk  to  a  disk:  An 
exact  solution,  J.  Acoust.  Soc.  Am.  61 ,  318- 
324  (1977). 


-(Z2  +   R2   +  Q2) 


.(Z)  =  ^  e 


2a2 


Z(R2  +  Q2) 

0  0 


[2]    Rhyne,  T.  L. ,  Acoustic  Instrumentation  and 
Characterization  of  Lung  Tissue  (Forest 
Grove,  Oregon,  1977). 

[3]    Rhyne,  T.  L. ,  Sonar  Characterization  of  Tis- 
sue as  Applied  to  the  Lung,  Sc.  D.  Thesis, 
Department  of  Electrical  Engineering  and  Com- 
puter Science,  Massachusetts  Institute  of 
Technology,  Cambridge,  Massachusetts  (1976). 


Z  >  0 


(B-5) 


-(R^  +  Q^) 


Z  =  e 


4a2       a^Tt  ' 


R2  +   Q2\  R2  +  Q2 

1  +  _o  ^1  i_       0  ^0 


2t2 


4t2 


R2  +  Q2        /r2   +  q2 


2t2 


4t2 


(B-6) 


j2  = 
z 


Z2 


-7) 


The  mean  squared  lung  surface  reflectance  in 
eq.  (B-8)  expresses  the  energy  scattered  at  a 
given  frequency.    This  function  is  essentially 
the  scattering  function  discussed  in  radar. 


|G(f)| 


2   =   ,2  = 


R2  .  Q2  .  2a2 


(B-E 


Acknowledgments 


This  paper  is  essentially  a  summary  of  work  re- 
ported in  the  three  references  cited.    It  was  felt 
that  it  would  be  useful  to  the  reader  to  report 
the  essential  results  here  in  summary  form  and  to 
refer  the  interested  reader  to  the  more  extensive 
discussions  and  numerous  literature  citations  to 
be  found  in  the  basic  papers. 

The  work  summarized  here  and  in  the  basic 
papers  was  supported  by  the  Ambrose  Monell  Founda- 
tion and  the  G.  Linger  Vetlesen  Foundation. 


141 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,   Spec.   Publ.    525    (U.S.  Government  Printing  Office,  Washington,   D.C.,  1979). 


ANGLE  SCAN  AND  FREQUENCY-SWEPT  ULTRASONIC  SCATTERING  CHARACTERIZATION  OF  TISSUE 


R.  C.  Waag,  P.  P.  K.  Lee,  R.  M.  Lerner,  L.  P.  Hunter,  R.  Gramiak,  and  E.  A.  Schenk 

Departments  of  Electrical  Engineering,  Radiology,  and  Pathology 
University  of  Rochester 
Rochester,  New  York    14627,  U.S.A. 


Ultrasonic  wave  interference  has  been  applied  to  characterize  tissue  by  measuring 
scattered  wave  intensity  as  a  function  of  frequency  and  angle.    A  theory  of  acoustic 
wave  propagation  and  scattering  in  an  inhomogeneous  medium  has  been  employed  to  show 
that  received  ultrasound  may  be  expressed  in  terms  of  medium  refractive  index  varia- 
tions via  a  Fourier  transform.    Using  a  computer-based  system,  measurements  were  made 
on  model  targets  and  post-mortem  liver  specimens.    Model  studies  demonstrate  that 
regular  scatterer  spacing  can  be  inferred  from  measured  diffraction  data  by  Fourier 
inversion  and  that  scattering  differences  can  be  observed  from  random  media  consist- 
ing of  particles  of  different  average  sizes.    Scattering  from  liver  indicates  there 
is  significantly  more  energy  scattered  at  small  angles  than  is  backscattered.  Cor- 
relation of  ultrasound  scattering  with  structure  observed  through  a  microscope  has 
been  obtained  by  computing  the  diffraction  pattern  of  the  two-dimensional  optical 
transmittance  images  acquired  through  a  microscope-TV  chain  attached  to  the  computer. 
Average  particle  sizes  of  the  model  random  media  determined  by  Fourier  analysis  that 
exhibited  diffraction  rings  of  the  digitized  cross-sections  yielded  scattering  predic- 
tions which  were  in  agreement  with  measured  acoustic  data. 

Key  words:   Angular  scattering;  characterization  of  tissue;  optical  correlation; 
scattering;  swept-f requency  diffraction. 


1.  Introduction 

Although  ultrasound  has  been  widely  accepted 
as  a  valuable  diagnostic  tool  in  medical  imaging, 
its  use  in  quantitative  characterization  of  tis- 
sue is  just  beginning  to  emerge.    Among  methods 
which  employ  amplitude,  phase,  frequency,  and  at- 
tenuation of  ultrasound  signals,  diffraction- 
based  techniques  offer  the  potential  of  charac- 
terizing acoustic  scattering  element  distribu- 
tion on  a  scale  corresponding  to  the  wavelength 
within  a  finite  volume  probed  by  the  ultrasound 
beam  [l-6]i.    Success  of  wave  interference  methods 
in  x-ray  studies  of  materials  as  well  as  in  atmos- 
pheric probing  by  radar  motivates  the  development 
of  the  methodology  for  acoustic  studies  of  tissue. 
However,  realization  of  practical  tissue  charac- 
terization systems  based  on  diffraction  require 
more  detailed  understanding  of  the  acoustic  scat- 
tering process,  including  development  of  useful 
models  and  knowledge  of  practical  measurement 
limitations. 

In  this  paper,  our  ultrasonic  scattering  studies 
of  model  targets  and  preliminary  results  from  liver 
tissue  are  described.    A  model  is  developed  to  show 
the  fundamental  relationship  between  scattered 
acoustic  waves  and  tissue  structure.    The  measure- 
ments demonstrate  feasibility  of  the  method  and 
identify  problems.    The  results  indicate  the  prom- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


ise  of  the  concept  but  point  out  the  need  for  ad- 
ditional data  to  characterize  tissue. 

Our  research  combining  angle  scanning  and  fre- 
quency sweeping  is  intended  to  provide  a  founda- 
tion for  characterization  of  tissue  by  its  ultra- 
sonic scattering  properties. 

2.    Theoretical  Basis 

A.    Acoustic  Scattering  Model 

The  model  we  use  relates  the  scattered  intensi- 
ty to  the  acoustic  structure  of  the  tissue  and  in- 
dicates that  different  levels  of  organization  may 
be  studied  with  ultrasound  by  appropriate  choice 
of  frequency  and  geometry.    The  resulting  acoustic 
characterization  can  then  be  compared  to  micro- 
scopic determinations  of  structure  to  establish 
tissue  pathology.    The  theory  is  analogous  to  that 
used  in  x-ray  diffraction  of  amorphous  materials 
and  allows  the  structure  of  the  medium  to  be  in- 
ferred from  measurements  of  the  scattered  acoustic 
wave. 

The  tissue  is  modeled  as  an  inhomogeneous  medium 
in  which  the  acoustic  properties  exhibit  variations 
from  point  to  point.    Since  wave  propagation  is  af- 
fected by  the  local  environment,  wavefronts  are  dis- 
torted and  scattered.    The  derivation  has  followed 
the  basic  approach  used  in  modeling  wave  propaga- 
tion and  scattering  for  acoustic  studies  of  the 
lower  atmosphere  [7]. 


143 


The  analysis  begins  with  consideration  of  a 
plane  wave  propagating  through  a  lossless  unbounded 
medium  which  has  small  variations  in  refractive 
index.    Small  values  of  absorption  do  not  change 
the  basic  relationships  [6]. 

Assuming  a  monochromatic  incident  plane  wave, 
weak  single  scattering,  and  a  receiver  in  the  far- 
field,  the  resulting  scattered  wave  velocity  poten- 
tial 't'llrjk)  is  given  by 


(t.i(r,k)  = 


A  k2  e^^^    r  ■ 


k-r' 


2-nr 


ni(r' )dV' 


(1) 


V 


in  which 

Aq  =  incident  plane  wave  amplitude 
k    =  incident  wave  number 
K    =  scattering  wave  vector 
ni(r')  =  variations  of  acoustic  refractive  index 

V  =  scattering  volume 
r    =  distance  from  the  scattering  volume 
center  to  the  reception  point 

Thus,  the  velocity  potential  at  a  receiving  point  r 
is  equal  to  the  product  of  a  frequency-dependent 
factor  and  the  three-dimensional  spatial  Fourier 
transform  of  the  acoustic  refractive  index  varia- 
tions. 

The  theory  also  relates  the  mean  square  value 
of  velocity  potential  to  the  correlation  function 
of  the  acoustic  refractive  index  variations  in  a 
random  medium  if  the  variations  are  stationary  in 
space.    The  relation  is 


:|4>i(r,k)|: 


A2k'*V' 

0 


"1 


(p)dV' 


(2) 


where 


(p)  =  <ni(r' )ni(r'  +  p)> 


where  <  >  denotes  ensemble  averaging.    This  shows 
that  the  average  intensity  of  the  scattered  wave 
is  proportional  to  the  three-dimensional  spatial 
Fourier  transform  of  the  correlation  function  of 
the  acoustic  refractive  index  variations. 

The  expression  for  the  mean  square  velocity 
potential  allows  determination  of  the  correlation 
function  of  the  acoustic  refractive  index  varia- 
tions in  the  medium  by  inversion  of  the  transform 
relating  tbe  measured  average  scattered  intensity 
as  a  function  of  scattering  vector  [8].  Alterna- 
tively, a  specific  form  for  the  correlation  func- 
tion can  be  employed  to  evaluate  the  average  scat- 
tered intensity  for  comparison  with  experimental 
data.    The  latter  approach  has  been  used  to  evalu- 
ate scattering  from  a  cloud  of  scatterers  assuming 
a  correlation  function  with  a  Gaussian  shape  modi- 
fied to  remove  the  dc  spectral  component  [9]. 
Neglecting  density  variations,  the  resultant  ex- 
pression for  the  scattered  intensity  per  unit  in- 
cident intensity  is 


(3) 


k2a2sin2|)e 


-2k2a^ 


.2^ 


V  =  scattering  angle 

NV  =  mean  number  of  scatterers  in  sample  volume 
a  =  mean  radius  of  the  particles 
=  mean  value  of  compressibility  variations 
between  scatterers  and  medium 

The  result  predicts  that  smaller  particles  scatter 
more  energy  off-axis  than  large  particles  for  a 
fixed  incident  frequency. 

B.    Optical  Correlation 

Optical  data  for  comparison  to  ultrasound  angle 
scattering  data  can  be  obtained  by  computing  the 
diffraction  pattern  of  a  sample  region  with  vari- 
able transmission  representing  spatial  distribu- 
tion of  scattering  inhomogeneities.    In  the  far- 
field  of  a  plane  through  the  region,  the  diffracted 
optical  field  is  given  in  terms  of  the  two-dimen- 
sional Fourier  transform  of  the  transmission  pat- 
tern [10]. 


Ae 


-i  kr 


J  ^^l  >^  "  - 


dr' 


(4) 


where 


A 
k 

T(r') 


constant 

2i  r 
\  r 

transmittance  of  aperture  S 


The  intensity  of  I(r)  of  the  pattern  is  then 

I(r)  =  *(r)**(r) 

The  dimensions  over  which  diffraction  occurs  are 
determined  by  the  wavelength  or,  equivalently,  the 
sampling  interval  of  the  optical  transmittance  and 
can  be  readily  chosen  in  analysis  to  yield  a  scale 
comparable  to  that  in  ultrasound  diffraction. 
Averaging  the  two-dimensional  diffraction  intensity 
patterns  of  any  collection  of  typical  cross-sections 
in  the  region  results  in  a  statistical  characteri- 
zation of  diffraction  by  a  random  medium. 

For  an  isotropic  random  medium  with  local  order, 
the  predicted  average  diffracted  intensity  takes 
the  form  of  concentric  rings  [ill.    The  spacing  of 
the  rings  gives  nearest  neighbor  distance  while  the 
width  of  the  rings  is  a  measure  of  the  spread  in 
scatterer  size  and  the  number  of  visible  rings  in- 
dicates ranges  of  local  order.    Optical  diffraction 
also  allows  quantitative  determinations  of  param- 
eters describing  known  optical  morphology  for  com- 
parison with  similar  parameters  derived  ultrasoni- 
cally  when  there  is  no  order  since  the  width  or 
general  shape  of  the  power  spectra  can  be  used  in 
these  cases. 

3.    Experimental  Methods 

A.    Acoustic  Scattering 

Initial  studies  employed  frequency  sweeping 
with  fixed  transducer-target  orientation.  The 
scattering  experiments  were  performed  in  a  water- 
filled  tank  containing  a  fixture  on  which  trans- 
ducer holders  were  mounted.    The  holders  could  be 
angled  to  align  the  ultrasonic  beams  and  could  be 
slid  along  the  fixture  as  well. 


144 


The  electronic  instrumentation  used  for  this 
investigation  consisted  of  a  transmitter  capable 
of  frequency  sweeping,  a  range-gated  receiver  and 
a  recorder.    Bursts  of  ultrasound  were  gated  from 
a  variable  frequency  master  oscillator  to  a  power 
amplifier  which  drives  a  wideband  9.5  mm  diameter 
transducer  with  a  5  MHz  center  frequency.  The 
frequency  of  the  master  oscillator  was  slowly 
swept  over  a  range  of  2  to  8  MHz  by  a  linear  ramp 
function  that  produced  negligible  carrier  frequen- 
cy variation  over  any  single  pulse. 

The  scattered  ultrasound  pressure  was  detected 
by  a  receiving  transducer  matched  to  that  used  for 
transmission  and  the  amplified  RF  signal  was 
energy-detected  to  give  a  video  signal  proportion- 
al to  the  incident  wave  peak  intensity  which  was 
then  recorded. 

Arrays  of  cylinders  were  studied  to  provide  in- 
formation about  the  influence  of  geometric  align- 
ment and  beam  uniformity.    The  scatterer  positions 
for  array  measurements  were  established  by  five 
sets  of  uniformly  spaced  grooves  two  centimeters 
in  length.    A  rubber  sheet  was  used  to  shield  the 
holder  exposing  only  one  set  of  wires  for  acoustic 
measurements.    The  center  of  the  array  was  nominal- 
ly set  at  the  intersection  of  the  transducer  beams 
for  each  study  with  the  array  elements  lying  along 
the  bisector  of  the  angle  formed  by  the  beams. 

The  energy  detector  output  was  bandpass  filter- 
ed to  produce  a  signal  that  was  displayed  as  a 
function  of  frequency  on  the  face  of  an  oscillo- 
scope for  photographic  recording.    The  analog  curv- 
es were  sampled  at  0.115  MHz  intervals  and  quan- 
tized into  integers  in  the  range  0  to  255  for  cal- 
culation of  Fourier  transforms. 

Ultrasound  scattering  as  a  function  of  angle 
was  measured  in  a  water  tank  using  an  acoustic 
diffractometer  [12]  that  provided  precise  control 
of  transducer  position  relative  to  the  scattering 
volume.    The  scattering  medium  was  contained  in  a 
cylindrical  column  28  mm  in  diameter  by  a  .0254 
mm  thick  cellulose  tubing  that  minimized  reflected 
energy  loss  at  the  perimeter  of  the  sample.  Trans- 
mitting and  receiving  transducers  containing  4.76 
mm  radius  ceramic  disks  as  active  elements  were 
rotated  in  equal  increments  but  opposite  directions 
about  the  axis  of  the  cylindrical  sample.  Trans- 
mitter and  receiver  distances  from  the  axis  of 
rotation  were  each  equal  to  13.6  cm.    This  scanning 
procedure  maintains  a  constant  scattering  vector 
direction  while  changing  the  scattering  angle,  v, 
and  hence  the  scattering  vector  magnitude.  The 
centers  of  transducer  rotation  were  maintained  to 
within  ±  0.3  mm  during  the  angle  scan. 

The  electronics  were  the  same  as  in  the  swept- 
frequency  measurements.    The  receiver  electronics 
were  also  the  same  as  used  in  the  previous  studies 
except  that  a  log  amplifier  was  added  to  accommo- 
date both  the  small  off-axis  scattering  and  the 
large  direct  transmission  signals.    The  energy  de- 
tected output  was  sampled  by  the  computer  and  re- 
corded in  digital  form  along  with  appropriate  posi- 
tional information  for  statistical  analysis  and 
display. 

A  typical  scattering  experiment  consisted  of 
sixteen  angle  scans  taken  at  2.5  mm  increments 
along  the  axis  of  the  cylindrical  sample.    In  each 
angle  scan,  data  was  collected  at  163  equally 
spaced  increments  of  2°  from  a  scattering  angle  of 
-153°  to  165°.    The  transmitted  tone  burst  was 
10  MS  long  while  the  receiver  gate  was  6  ys  long 
and  centered  in  the  scattering  volume. 


Fresh  samples  of  pig  liver  and  human  liver 
samples  obtained  at  autopsy  have  also  been  studied 
ul trasonical ly  by  angle  scanning  and  varying  fre- 
quency.   Cylindrical  plugs  were  taken  from  peri- 
pheral regions  near  the  capsule  to  minimize  the 
inclusion  of  large  vessels.    Transmit  pulse  length 
and  receiver  gates  were  set  as  in  the  random  medi- 
um model  studies  to  minimize  inclusion  of  specular 
reflectors  from  the  surface. 

B.    Optical  Data 

Optical  data  was  acquired  under  computer  con- 
trol by  sampling  a  TV  image  obtained  through  a 
transmission  microscope.    Drops  of  the  particle 
suspensions  were  spread  over  a  microscope  slide 
and  typical  areas  digitized.    Five  different 
images  from  a  smear  of  each  particle  suspension 
were  obtained  for  analysis.    The  individual  images 
consisted  of  8-bit  samples  representing  light  in- 
tensity transmission  over  a  field  of  1.7  mm  span- 
ned by  384  x  384  matrix  elements. 

The  cylindrical  liver  specimens  were  fixed  in 
formalin  and  standard  sectioning  and  staining 
procedures  used  to  evaluate  histomorphologic  chang- 
es in  the  sections  that  were  studied  ultrasonical- 
ly.    Each  stained  section  was  digitized  into  a 
256  x  256  matrix  covering  a  field  of  4.5  mm.  In 
the  optical  studies  of  liver  tissue,  a  Van  Gieson 
connective  tissue  stain  was  employed  to  bring  out 
structures  believed  to  cause  ultrasonic  scattering 
and  the  computed  average  power  spectra  were  used 
to  characterize  the  architecture  for  comparison 
with  ultrasound  angle  scan  data. 

The  digital  representations  of  the  optical 
images  obtained  through  the  microscope-TV  system 
were  enhanced  to  improve  the  contrast  between  the 
particle  boundaries  and  the  background.    This  was 
accomplished  by  forming  histograms  of  original 
image  amplitudes  and  using  these  histograms  to  de- 
velop mappings  that  spread  the  original  light  in- 
tensity transmission  values  over  wider  ranges. 
The  far-field  diffraction  pattern  of  each  optical 
image  was  obtained  by  a  two-dimensional  Fourier 
transformation  implemented  using  an  integer  FFT 
algorithm.    The  log  magnitude  of  the  resultant 
spectra  was  calculated  to  facilitate  display  of  a 
wide  range  of  spatial  frequency  amplitudes.  Aver- 
ages of  the  log  magnitude  spectra  from  the  five 
sample  functions  of  the  same  particle  size  distri- 
bution were  computed.    The  spherical  symmetry  of 
this  spectral  estimate  was  used  to  produce  addi- 
tional smoothing  by  averaging  spatial  frequency 
amplitudes  determined  along  radii  spaced  1°  apart. 
The  results  were  used  to  find  average  nearest 
neighbor  distances. 

4.  Results 

Data  was  collected  by  frequency  scanning  10  mil 
monofilament  nylon  cylinders  in  arrays  with  spac- 
ings  of  0.72  and  1.52  mm  at  a  range  of  11.0  cm 
and  a  Fourier  transform  was  applied  to  infer  the 
spacings  (fig.  1).    The  influence  of  transmitter 
range  on  measurements  of  diffraction  by  an  array 
spacing  of  1.52  mm  was  observed  for  ranges  of  4.5 
and  11.5  cm  (fig.  2).    The  distances  (0.78  and 
1.55  mm)  indicated  by  the  peaks  on  the  transformed 
data  in  each  of  the  measurements  are  within  4  per- 
cent of  the  actual  values.    The  envelope  of  the 
data  collected  at  4.5  cm  is  different  from  the  far- 
field  measurements,  but  did  not  invalidate  calcula- 


145 


Fig.  1.    Swept- Frequency  Diffraction  by  Arrays.    Measured  values  of  scat- 
tered energy  as  a  function  of  frequency  are  shown  (left)  on  a 
linear  scale  with  their  corresponding  Fourier  transforms  (right) 
also  on  a  linear  scale  for  array  spacings  of  1.52  mm  (top)  and 
0.76  mm  (bottom)  at  a  transmitter  range  of  11.0  cm.    The  ultra- 
sonically  determined  spacings  were  1.55  mm  and  0.78  mm  respec- 
tively. 


Fig.  2.    Swept-Frequency  Diffraction  as  a  Function  of  Range.  Measured 

curves  of  energy  as  a  function  of  frequency  are  displayed  (left) 
on  a  linear  scale  with  corresponding  Fourier  transforms  (right) 
also  on  a  linear  scale  for  transmitter  ranges  of  4.5  cm  (top) 
and  11.5  cm  (bottom)  for  an  array  spacing  of  1.52  tom.    The  ultra- 
sonically  determined  spacings  were  1.48  mm  at  the  far  range  and 
1.55  mm  at  the  near  range. 


146 


Fig.  3.    Frequency-Dependent  Angular  Scattering  by  a  Random  Medium  Model. 

Measurements  of  average  intensity  are  shown  (left)  on  a  logarith- 
mic scale  along  with  calculated  scattering  (right)  on  the  same 
scale  for  a  frequency  of  3.8  MHz  (top)  and  6.0  MHz  (bottom).  The 
measured  data  obtained  from  a  suspension  of  particles  demonstrates 
more  forward  scattering  as  frequency  increases  and  compares  well 
with  computations  of  scattering  from  a  cloud  in  which  the  average 
particle  radius  is  128  pm. 


tions  of  spacings  by  Fourier  inversion  of  the  data. 

Angle  scans  were  carried  out  on  three  suspensions 
of  different  average  size  Sephadex2  particles.  Mea- 
surements of  scattered  ultrasound  signal  as  a  func- 
tion of  angle  were  averaged  for  each  distribution 
of  particle  size.    The  averaged  angle  scan  data  is 
an  estimate  of  the  spectral  power  in  wave  number 
or  reciprocal  space  of  the  variations  in  acoustic 
index  of  refraction. 

Averages  of  ten  angle  scans  for  each  of  three 
particle  size  distributions  were  obtained.  The 
transmitted  main  beam  was  subtracted  from  the  aver- 
age data  and  the  resulting  angular  scattering  data 
smoothed  to  remove  artificial  fluctuations  arising 
from  averaging  only  a  finite  number  of  sample  func- 
tions.   The  final  result  was  compared  with  calcula- 


^Trademark  of  Pharmacia  Fine  Chemicals. 


tions  of  scattered  intensity  using  eq.  (3)  to  de- 
termine average  scattered  intensity  as  a  function 
of  angle  at  two  frequencies  for  the  coarse  grade 
of  particles  which  had  the  largest  average  size 
(fig.  3).    The  averages  for  two  other  particle 
distributions  show  how  off-axis  scattering  increas- 
es as  average  particle  size  decreases  (fig.  4). 

Enhanced  optical  transmittance  images  of  five 
sample  functions  and  the  corresponding  average  log 
magnitude  spectra  were  obtained  from  smears  of 
three  particle  grades  (fig.  5).    A  smoothed  spatial 
frequency  amplitude  distribution  was  found  by  aver- 
aging the  amplitudes  along  radii  from  which  the 
nearest  neighbor  distances  may  be  determined 
(table  1). 

Pilot  studies  of  acoustic  scattering  from  human 
and  pig  liver  tissue  by  angle  scanning  have  been 
carried  out  (fig.  6).    The  data  shows  that  there 


147 


Fig.  4.    Size-Dependent  Angular  Scattering  by  a  Random  Medium  Model.  The 
polar  plots  and  corresponding  Cartesian  displays  show  average  in- 
tensity on  a  linear  scale  at  5.9  MHz  for  distributions  of  scatter- 
ers  with  an  average  particle  radius  of  93  ym  (top)  and  with  an 
average  radius  of  84  m  (bottom).    The  increase  in  omnidirectional 
scattering  for  smaller  scatterers  is  evident  in  both  plots  and 
also  demonstrated  by  the  mean  scattering  angle  defined  by  arrows 
crossing  standard  deviation  bars  below  the  Cartesian  plots. 


Table  1.    Nearest  neighbor  spacing  determined 
by  Fourier  transform  analysis  of 
optical  transmi ttance. 

Particle         Mean  spacing         Spread  of 
distribution  (pm)  spacings  (+la) 

(ym) 


G50M  93.3  63.2  -  177.6 

G50F  84.3  57.3  -  159.4 


is  significantly  more  scattered  energy  at  small 
angles  than  in  the  backscattered  mode.  Scattering 
In  the  forward  direction  becomes  greater  as  the 
frequency  increases.    Optical  data  obtained  from 
these  specimens  shows  the  well-defined  regular 
lobular  structure  of  pig  liver  and  the  irregular 
structure  of  a  post-necrotic  cirrhotic  human  liver 
(fig.  7).    The  power  spectra  and  their  correspond- 
ing average  radial  distributions  show  the  same 
trend  of  increased  omnidirectional  scattering  from 
pig  liver  that  the  acoustic  data  demonstrate. 

5.  Discussion 

The  measured  scattering  from  arrays  of  nylon 
cylinders  yields  element  spacings  that  correspond 
well  to  actual  values  and  shows  the  influence  of 
near-field  variations  of  transducer  beams  on  the 


determination  of  scatterer  spacing.    The  number  of 
peaks  in  the  transformed  data  indicates  diffraction 
by  integer  multiples  of  the  basic  spacing  and  are  a 
measure  of  the  number  of  scatterers  in  the  beam  or 
the  effective  width  of  the  common  scattering  volume 
which  contains  more  array  elements  as  the  spacing 
decreases. 

Diffraction-based  measurements  of  array  element 
spacing  as  a  function  of  range  from  the  transmitter 
show  peaks  that  do  not  vary  in  position  when  the 
scatterers  are  in  the  near-field  of  the  trans- 
ducers.   These  measurements  also  support  the  utili- 
ty of  a  model  that  employs  the  far-field  assumption 
for  inferring  structure  even  when  the  far-field  as- 
sumption is  not  strictly  satisfied. 

Acoustic  scattering  from  Sephadex  was  measured 
using  a  wavelength  that  ranged  from  2  to  30  times 
the  particle  diameters  and  corresponds  to  a  transi- 
tion region  between  point  scattering  and  diffrac- 
tion by  individual  spheres.    This  relationship  of 
wavelength  and  particle  size  was  chosen  for  its  poten- 
tial similarity  to  situations  of  interest  in  medical 
ultrasound.    Differences  in  scattering  due  to  par- 
ticle size  are  still  demonstrated  by  the  analysis 
of  both  the  acoustic  and  optical  data.    The  acoustic 
data  from  the  Sephadex  experiments  agrees  with  the 
predictions  of  the  expression  describing  scattering 
from  a  cloud  of  particles. 

For  the  coarse  grade  of  particles,  the  computed 
results  agree  well  with  the  experimental  data. 
The  mean  value  size  parameter  of  128  ym  is  well 


148 


Fig.  5.    Optical  Image  Analysis  of  Random  Media  Model.    In  (a)  particles 
with  an  average  size  of  84  pm  (left)  and  93  ym  are  shown  with 
the  central  field  enhanced  by  amplitude  mapping.    The  power 
spectra  (b)  obtained  by  averaging  showed  diffraction  rings  with 
an  average  radial  distribution  of  energy  (c)  that  is  inversely 
proportional  to  scatterer  size  and  consistent  with  the  acous- 
tic scattering  results. 


within  the  range  of  the  manufacturer's  specifica- 
tions on  sizes  and  is  also  confirmed  by  the  optical 
determinations.    Both  experimental  data  and  calcula- 
tions show  that  the  scattering  becomes  greater  in 
the  forward  direction  as  frequency  increases.  Polar 
plots  of  average  acoustic  intensity  as  a  function 
of  scattering  angle  also  show  that  a  distribution 
of  small  particles  scatters  relatively  more  energy 
at  larger  angles  than  does  a  distribution  of  large 
particles. 

These  results  are  in  agreement  with  theoretical 
predictions  about  scattering  off-axis  as  a  function 
of  particle  size,  and  support  the  swept-frequency 
studies  in  which  small  particles  scattered  more 
energy  at  high  frequency  than  did  large  particles 


reported  earlier  [4].    The  experiments  are  compar- 
able because  the  Sephadex  suspensions  are  isotropic 
so  that  angle  scanning  produces  the  same  result  as 
frequency  scanning. 

The  acoustic  scattering  data  was  collected  under 
conditions  that  reasonably  approximate  the  plane 
wave,  point  receiver,  and  weak  scattering  assump- 
tions employed  in  the  theoretical  development.  The 
170  microseconds  between  transmit  and  receive  gates 
correspond  to  transducer  distances  of  12.6  cm  from 
the  center  of  the  scattering  volume.    This  distance 
normalized  by  the  ratio,  the  square  of  transducer 
radius  to  the  wavelength,  is  10.9  and  is  well  with- 
in the  far-field  resulting  in  a  plane  wave  incident 
on  the  scattering  volume.    The  distance  also  results 


149 


Fig.  6.    Comparison  of  Average  Angular  Scattered  Intensity  From  Normal  Pig 
Liver  and  Cirrhotic  Human  Liver.    Specimens  studied  at  3  MHz  (upper 
panels)  show  that  the  normal  pig  liver  scatters  more  omnidirectionally 
than  cirrhotic  human  liver  and  that,  as  the  frequency  is  increased  to 
6  MHz,  the  forward  scattering  component  increases  in  both  cases.  This 
is  consistent  with  scattering  from  collagen  which  is  arranged  in  a 
smaller,  more  uniform  matrix  in  the  pig  liver  than  in  the  cirrhotic 
human  liver. 


in  a  maximum  phase  variation  of  16.4°  across  the 
face  of  the  receiving  transducer  for  a  wave  origi- 
nating from  a  point  at  the  center  of  the  scattering 
volume  and  thus  a  point  receiver  approximation  is 
justified.  . 

The  logarithmic  compression  of  the  received 
ultrasound  signal  near  0°  scattering  angle,  needed 
to  allow  simultaneous  demonstration  of  scattering 
signals  at  other  angles,  indicates  that  the  assump- 
tion of  weak,  single  scattering  or  the  Born  ap- 
proximation is  valid. 

The  polar  plots  do  not  demonstrate  symmetric 
scattering  about  each  side  of  the  0°  scattering 
angle  beam  as  anticipated  from  the  isotropy  as- 
sumption. A  reason  for  this  is  that  a  slightly 
different  scattering  volume  was  sampled  on  each 
side  of  0°  due  to  error  in  concentricity  of  the 
transducer  arms. 

Optical  characterization  of  a  distribution  of 
spheres  shows  a  ring  which  agrees  with  theory. 
The  good  agreement  between  theory  and  final  ring 
structures  obtained  by  processing  only  five  samples 


supports  the  notion  that  the  particle  distribution 
used  for  a  random  medium  model  is  isotropic.  Since 
the  particles  are  closely  packed,  the  spacings  ob- 
served by  light  transmittance  analysis  are  equal 
to  particle  size. 

Parameters  for  the  particle  distributions  de- 
rived from  averages  of  radial  lines  in  the  two- 
dimensional  log  spectra  are  in  excellent  agree- 
ment with  the  values  supplied  by  the  manufacturer 
as  well  as  being  in  qualitative  agreement  with 
analysis  of  acoustic  scattering  data. 

The  angle  scan  data  from  pig  and  human  liver 
specimens  demonstrate  differences  as  a  result  of 
the  different  architecture  of  the  tissues.  The 
pig  liver  scatters  more  omnidirectionally  than  the 
human  liver.    At  lower  frequencies,  the  pig  liver 
also  shows  a  lobe  structure  indicative  of  local 
order  in  the  lobular  arrangement,  while  the  human 
liver  scattering,  which  is  predominantly  in  the 
forward  direction,  results  from  larger,  more  ir- 
regular spacings.    The  increase  in  forward  scat- 
tering at  high  frequencies  as  in  the  Sephadex  ex^ 


150 


Fig.  7.    Optical  Image  Analysis  of  Liver.    In  (a),  normal  pig  liver  (left)  and 
cirrhotic  human  liver  have  been  stained  to  show  connective  tissue  as 
dark  bands.    The  power  spectra  (b)  obtained  by  averaging  show  that  the 
pig  liver  produces  a  broader  distribution  of  scattering  than  the  human 
liver.    The  relatively  isotropic  patterns  have  been  reduced  to  a  one- 
dimensional  form  (c)  by  averaging  radial  lines  at  1°  increments  to  produce 
data  comparable  to  ultrasound  angular  scattering  when  viewed  as  a  Carte- 
sian plot.    The  optical  analysis  demonstrates  the  same  increased  spread 
in  scattering  from  pig  liver  that  the  acoustic  patterns  contain. 


151 


periments  indicates  that  important  scattering  may 
take  place  from  elements  in  the  size  range  includ- 
ed in  the  Sephadex  studies.    The  data  also  implies 
that  backscattered  energy,  while  easier  to  monitor 
in  practical  situations,  represents  only  a  small 
amount  of  tissue  structure  information.    It  re- 
mains to  determine  whether  observed  peaks  in  the 
scattering  pattern  are  due  to  many  randomly  spaced 
scatterers  in  the  tissue  or  a  few  large  single 
scattering  objects. 

Further  investigation  is  also  required  to  deter- 
mine the  specific  structural  changes  observed 
acoustically,  correlate  these  results  with  histo- 
pathology,  and  establish  the  reproducibility  of 
the  results  from  these  initial  studies  of  liver 
specimens . 

6.  Conclusion 

Experimental  observation  of  acoustic  diffraction 
by  arrays  using  swept-frequency  ultrasound  have 
agreed  well  with  theoretical  predictions.    The  ex- 
periments indicate  that  near-field  beam  nonuniformi- 
ties  need  not  invalidate  inference  of  scatterer 
spacing  from  a  model  employing  a  far-field  approxi- 
mation.   However,  more  results  are  needed  to  bound 
limitations  imposed  on  measurement  of  scattering 
by  acoustic  diffraction  as  a  result  of  beam  non- 
uniformities  and  misalignments. 

The  ultrasound  angle  scan  data  from  Sephadex 
shows  size-dependent  scattering  and  agrees  with 
theoretical  predictions  as  well  as  with  optical 
analysis.    The  Fourier  analysis  of  optical  data 
accurately  confirms  the  particle  size  distribution 
information  provided  by  the  vendor  and  demonstrat- 
es a  way  optical  images  can  be  quantitatively 
analyzed  to  provide  a  reference  for  comparison  with 
ultrasound  data.    The  results  obtained  indicate  the 
potential  for  remote  probing  with  ultrasound  to 
characterize  a  random  medium  from  its  scattering 
and  the  use  of  digitally  processed  optical  data 
to  confirm  the  results. 

Preliminary  data  from  swept-frequency  and  angle 
scan  experiments  with  pig  and  diseased  human  liver 
tissue  shows  the  importance  of  scattering  at  small 
angles  and  also  that  this  technique  may  lead  to  a 
useful  diagnostic  tool.    However,  additional  data 
is  needed  to  characterize  tissue  and  thus  point 
the  way  to  optimum  parameters  for  clinical  deter- 
minations of  tissue  structure  from  diffraction  ef- 
fects with  ultrasound. 

Acknowledgments 

We  wish  to  acknowledge  important  assistance  in 
the  conduct  of  this  work  by  Edward  E.  Eyler,  C. 
Robert  Hoffman  III,  and  Jeffrey  Astheimer  who  de- 
veloped computer  programs,  and  Frank  H.  Slaymaker 
and  Peter  H.  Helmers  who  designed  major  parts  of 
the  mechanical  and  electronic  hardware  used  in 
these  studies. 

This  work  was  supported  by  the  National  Science 
Foundation  under  grant  #APR75-14890  and  the  Nation- 
al Heart  and  Lung  Institute  under  grant  #HL15016. 


[2]    Chivers,  R.  C,  Hill,  R.  R. ,  and  Nicholas,  D,, 
Frequency  Dependence  of  Ultrasonic  Backscatter- 
ing  Cross-Section:    an  Indicator  of  Tissue 
Structure  Characteristics,  in  Proc.  of  the  2nd 
World  Congress  on  Ultrasonics  in  Medicine, 
Rotterdam,  The  Netherlands,  4-8  June,  1973, 
M.  deVlieger,  D.  N.  White,  and  V.  R.  McCready, 
eds.,  pp.  300-303  (Excerpta  Medica,  Amsterdam, 
1974). 

[3]    Heyser,  R.  C.  and  LeCroissette,  D.  H.,  A  new 
ultrasonic  imaging  system  using  time  delay 
spectrometry.  Ultrasound  in  Medicine  and  Bi- 
ology 1,  (2),  119  (March  1974).  ~ 

[4]    Waag,  R.  C.  and  Lerner,  R.  M. ,  Tissue  Macro- 
Structure  Determination  with  Swept-Frequency 
Ultrasound,  in  Proc.  of  1973  Ultrasonics 
Symposium,  Monterey,  Calif.,  IEEE  Cat.  No. 
73  CHO  807-8  SU  63-66,  5-7  November,  1973. 

[5]    Namery,  J.  and  Lele,  P.  P.,  Ultrasonic  Detec- 
tion of  Myocardial  Infarction  in  Dog,  in 
Proc.    of  1972  Ultrasonics  Symposium,  Boston, 
Mass.,  IEEE  Cat.  No.  72  CHO  708-8  SU:  491- 
494,  4-7  October,  1972. 

[6]    Waag,  R.  C,  Lerner,  R.  M. ,  and  Gramiak,  R. , 
Swept-Frequency  Ultrasonic  Determination  of 
Tissue  Macrostructure,  in  Ultrasonic  Tissue 
Characterization,  M.  Linzer,  ed..  National 
Bureau  of  Standards  Spec.  Publ.  453,  pp.  213- 
228  (U.S.  Government  Printing  Office,  Washing- 
ton, D.C.,  1976). 

[7]  Tatars kii,  V.  I.,  Wave  Propagation  in  a  Tur- 
bulent Medium  (McGraw-Hill  Book  Company,  New 
York,  1961). 

[8]    Lee,  P.  P.  K.,  Waag,  R.  C,  and  Hunter,  L.  P., 
Swept-frequency  diffraction  of  ultrasound  by 
cylinders  and  arrays,  J.  Acoust.  Soc.  Am.  63 
(2),  600-606  (1978). 

[9]    Morse,  P.  M.  and  Ingard,  K.  U.,  Theoretical 
Acoustics,  Chapter  8,  p.  439  (McGraw-Hill , 
Inc. ,  New  York,  1968). 

[10]    Goodman,  J.  W.,  Introduction  to  Fourier  Op- 
tics (McGraw-Hill  Book  Company,  New  York, 
1968). 

[11]    Azaroff,  L.  V.  et  al.,  X-Ray  Diffraction 

(McGraw-Hill  Book  Company,  New  York,  1974). 

[12]    Slaymaker,  F.  H.,  Ultrssonic  diffraction  ap- 
paratus users  manual,  (Preliminary)  Progress 
Report  on  Tissue  Characterization  with  Ultra- 
sound (NSF  Project  No.  APR  75-14890),  Ap- 
pendix I,  Dept.    of  Electrical  Engineering, 
College  of  Engineering  and  Applied  Science, 
Rochester,  New  York,  June  1976.  NTIS 
#PB267397/AS. 


References 


[1]    Baum,  G.,  Quantized  Ultrasonography,  in  Recent 
Advances  in  Diagnostic  Ultrasound,  E.  Rand, 
ed.  (Charles  C.  Thomas,  Publisher,  Spring- 
field, 111.,  1971). 


152 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.   Linzer,   ed . ,  National  Bureau 
of  Standards,   Spec.   Publ.   525    (U.S.  Government  Printing  Office,  Washington,   D.C.,  1979). 


QUANTITATIVE  MEASUREMENTS  OF  SCATTERING  OF  ULTRASOUND  BY  HEART  AND  LIVER 


J.  M.  Reid  and  K.  K.  Shung 


Institute  of  Applied  Physiology  and  Medicine 
and 

Providence  Medical  Center 
Seattle,  Washington    98122,  U.S.A. 


Quantitative  backscattering  measurements  have  been  made  using  the  substitution  method 
used  to  measure  the  scattering  properties  of  blood  previously.    We  have  obtained  pre- 
liminary experimental  values  Tor  the  backscattering  coefficients  of  liver  and  myocardium, 
fully  corrected  for  equipment  and  transducer  parameters  as  well  as  attenuation  of  the 
scattering  tissue.    Results  show  that  scattering  for  both  liver  and  myocardium  is  an 
increasing  function  of  frequency  up  to  10  MHz.    This  increase  indicates  that  the  tissue 
elements  responsible  for  the  scattering  must  be  less  than  30  micrometers  in  diameter. 
The  level  of  scattering  from  liver  tissue  indicates  that  if  it  is  isotropic  scattering 
the  scattering  loss  is  lange  enough  to  account  for  about  28  percent  of  the  total  at- 
tenuation reported  at  10  MHz  frequency. 


Keywords:    Absorption;  attenuation;  heart;  liver;  scattering;  scattering  cross-section. 


1.  Introduction 

The  recent  interest  in  characterizing  biologi- 
cal tissues  through  noninvasive  ultrasonic  means 
has  resulted  in  investigations  and  reports  on 
values  of  velocity,  attenuation,  impedance,  and 
scattering  and  echo  strength.    The  first  three 
quantities  are  nearly  always  measured  in  a  quanti- 
tative sense,  the  last  two  are  usually  measured 
in  a  relative  sense.    The  last  two  quantities  are 
sometimes  spoken  of  interchangeably  as  if  they  were 
the  same  thing.    However,  echo  strength  is  the  re- 
sult of  the  operation  of  a  number  of  factors: 

1)  the  absolute  scattering  strength  of  the  tissues; 

2)  the  absorption  of  the  tissues  (including  the 
tissue  that  is  doing  the  scattering);  and  3)  many 
equipment  parameters. 

Although  it  may  appear  to  be  expeditious  to  do 
characterization  on  the  basis  of  echo  amplitudes 
produced  by  existing  equipment,  we  believe  it 
preferable  to  obtain  the  data  under  conditions 
which  allow  the  absolute  scattering  coefficients 
to  be  calculated  independent  of  tissue  absorption 
and  equipment  parameters.    Knowledge  of  absolute 
scattering  parameters  is  useful  for  the  identifica- 
tion of  mechanisms  and  the  prediction  of  possible 
applications  in  medicine.    An  even  more  exciting 
use,  however,  is  in  the  design  of  new  equipment 
utilizing  new  approaches  to  tissue  characteriza- 
tion which  can  generally  only  be  done  if  the  data 
obtained  are  independent  of  the  taking  equipment. 

Quantitative  backscattering  measurements  can  be 
made  using  a  substitution  method  which  has  already 
been  used  to  measure  the  scattering  properties  of 


blood  [l-4]i.    We  have  used  this  method  to  obtain 
preliminary  experimental  values  for  the  scattering 
properties  of  liver  and  myocardium.    Further  more 
extensive  studies  on  human  tissues  will  be  con- 
ducted.   Animal  tissues  were  used  for  coneenience 
in  adapting  the  system  to  solid  tissues. 

Results  have  been  presented  by  Nicholas  [5]  on 
backscattering  coefficients  per  solid  angle  for 
liver,  spleen  and  brain.    No  details  are  given  on 
the  formalism  used  to  derive  the  backscattering 
coefficient.    Nicholas'  values  for  liver  are  about 
an  order  of  magnitude  greater  than  our  values,  and 
at  the  higher  frequencies  would  predict  a  scatter- 
ing loss  which  equals  the  entire  observed  attenua- 
tion.   Since  this  a  relatively  new  field,  we  be- 
lieve it  essential  to  give  as  many  details  of  the 
measurement  and  data  reduction  process  as  possible 
and,  in  this  article,  will  present  our  complete 
procedure. 

2.  Method 

The  backscattering  coefficient  defined  as  power 
scattered  per  unit  solid  angle  in  the  backward 
direction  by  a  unit  volume  of  scatterers  was  cal- 
culated by  comparing  the  RMS  value  of  the  backscat- 
tered  wave  to  that  of  a  known  reflected  wave.  The 
coefficient  can  be  defined  to  be  independent  of 
the  measuring  system  and  the  attenuation  charac- 
teristics of  the  tissue  [1,2]. 

Measurements  were  performed  in  a  large  water 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


153 


tank  which  was  filled  with  normal  saline  solution. 
The  temperature  of  the  saline  was  kept  at  18  °C  + 
1  °C.    The  samples  were  fixed  on  a  sample  holder 
about  15  cm  away  from  the  transducer.  Excised 
calf  liver  and  heart  were  obtained  immediately 
after  the  sacrifice  of  the  animal  and  stored  in 
normal  saline  solution  at  a  temperature  of  4  to 
6  °C.    Measurements  were  carried  out  within  three 
days. 

Five  liver  specimens  and  five  heart  specimens 
were  used  with  each  cut  into  sample  regions  which 
were  examined  so  that  only  those  samples  free  of 
gross  connective  tissue  or  blood  vessel  interface 
were  used  for  the  measurements  to  better  define 
the  bulk  properties  of  each  tissue  type.  The 
samples  were  2  cm  x  2  cm  squares  with  a  thickness 
of  3  to  5  cm.    The  liver  samples  were  examined  in 
a  number  of  randomly  chosen  orientations  so  that 
30  to  45  experimental  values  were  calculated  for 
each  point  on  the  results.    In  the  measurements  of 
myocardium  samples,  the  direction  of  the  sound  beam 
was  always  perpendicular  to  the  muscle  fibers. 

Ultrasonic  transducers  (Panametrics )  with 
resonant  frequencies  at  3,  5  and  10  MHz  and  a  diam- 
eter of  0.635  cm  were  used.    Electronic  equipment 
consisted  of  a  gated  burst  generator,  pulse  echo 
receiver,  and  a  controllable  receiver  gating  cir- 
cuit [2,3].    All  time  measurements  were  made  on  an 
oscilloscope  and  the  output  gated  echo  measured  by 
a  true  RMS  voltmeter.    A  calibrated  attenuator,  im- 
pedance matched  to  the  receiver  input  circuit,  was 
adjusted  to  obtain  the  same  reference  reading  on 
the  voltmeter  for  the  reflected  wave  and  the  scat- 
tered wave. 

The  backscattering  coefficient  was  calculated 
from  the  attenuator  readings,  equipment  and  equa- 
tions expressing  the  received  available  power  from 
targets  located  on  the  axis  in  the  far  field  of  a 
round  piston  transducer  [1].  The  received  avail- 
able power  in  the  wave  reflected  from  a  plane  in- 
terface can  be  written 


where  nd  =  backscattering  coefficient  per  unit 
solid  angle,  per  cm^ 
S    =  cross-sectional  area  of  the  sound  beam, 
cm2 

o    =  attenuation  coefficient  of  the  scatter- 
ing tissue 

c    =  velocity  of  sound  in  the  scattering 
tissue 

R    =  range  to  surface  of  scattering  tissue 
ti  =  time  from  first  surface  echo  to  open- 
ing of  receiver  gate 
ta  =  time  from  first  surface  echo  to  closing 
of  receiver  gate 

Equation  (2)  neglects  the  front  surface  reflection 
of  the  sample  since  the  cut  specimens  we  employed 
did  not  include  surface  covering  membranes,  the 
usual  source  of  such  reflections.    The  terms  in 
brackets  are  correction  factors  to  account  for 
the  total  attenuation  losses  of  the  scattering 
material.    These  losses  include  absorption  of  the 
sound  in  the  scattering  tissue  between  its  sur- 
face and  the  region  from  which  scattered  echoes 
are  gated  into  our  receiver  and  measuring  circuit, 
absorption  of  sound  during  the  transmitted  pulse 
and  interaction  factors  which  depend  on  the  absorp- 
tion of  the  wave  during  the  duration  of  the  re- 
ceiver gate  and  the  transmitted  pulse.    The  scat- 
tering coefficient  can  be  found  by  dividing  eq. 
(2)  by  eq.  (1)  and  solving  for  the  coefficient. 
The  resulting  equation  is: 


E^A^xe 


.4amR 


0 


R2Tp4x2 


k2 


(1) 


where      =  received  available  electrical  power  (RMS) 
Pq  =  transmitter  electrical  power  (RMS) 
E    =  transducer  efficiency  (a  function  of 

frequency) 
A    =  transducer  area 
R    =  range  from  transducer  to  interface 
Tp  =  pulse  repetition  period 
T    =  transmitted  pulse  length 
X    =  wavelength 

am  =  pressure  attenuation  coefficient  for 

propagation  medium 
k    =  amplitude  reflection  coefficient  between 

propagating  medium  and  reflector 

In  a  similar  manner,  the  available  power  in  a  wave 
scattered  from  a  distribution  of  lossy  scatterers, 
Pg,  can  be  written  [2]: 


E2A2S2e 


.4amR 


P  8R'*TDX2a2c 
0  P 


(2) 


nd(e 


-2act 


aCT  _aCTK 
-  e  ) 


2k2R2a2cT 


(3) 


P    s(e-2°'cti  _  g-2act2)(gaCT  _  g-acx. 


The  ratio  of  P^  to  Pp  is  obtained  from  the 
change  of  the  input  attenuator  settings  necessary 
to  make  the  RMS  readings  of  the  pulse  from  the 
flat  reflector  equal  the  reading  of  the  gated, 
scattered  waves.    Theoretically,  this  ratio  should 
be  expressed  in  terms  of  the  open  circuit  trans- 
ducer voltages  due  to  the  scattered  and  the  re- 
flected waves.    When  the  attenuator  is  connected 
we  read  instead  a  loaded  or  closed  circuit  voltage. 
Since  we  use  the  same  transducer  and  receiver  im- 
pedance for  the  two  waves,  the  loaded  voltage  is 
less  than  the  open  circuit  voltage  by  a  complex 
frequency-dependent  factor  which  depends  on  the  im- 
pedances in  the  circuit.    This  factor  is  the  same 
for  the  scattered  wave  as  for  the  reflected  wave, 
however,  and  hence  divides  out  when  the  ratio  is 
taken.    Therefore,  the  ratio  is  the  same  for  the 
open-circuit  as  for  the  loaded  condition.  The 
other  factors  in  eq.  (3)  can  be  obtained  by  direct 
measurement  of  apparatus  dimensions,  oscilloscope 
time  settings  and  separate  measurements  of  attenua- 
tion and  velocity  within  the  scattering  tissue. 
In  this  investigation,  these  tissue  properties 
were  taken  from  the  published  literature  [6-8]. 
The  time  lapse  between  the  receiver  gate  and  the 
tissue  -  saline  surface  (tj  is  in  the  order  of  10 
to  20  \im.    Attenuation  losses  in  this  layer  as  well 
as  losses  in  the  sample  volume  are  corrected  for 
by  use  of  eq.  (3).    The  amplitude  reflection  coef- 
ficient between  the  saline  and  tissue  surface  was 
less  than  0.1  and  assumed  negligible. 

Figure  1  is  the  data  collection  sheet  filled 
out  by  the  experimenter  at  the  time  of  measurement 
for  one  specimen  at  one  frequency.    The  section  of 
the  sheet  before  the  calibration  run  consists  of 
equipment  parameters.    The  calibration  run  allows 


154 


Date     *  MARCH,  1977 


Data  Collection  Sheet  for  the  Scattering  Experiment 


Sign_ 


Specimen:       CALF  MYOCARDIUM 
Wave  Frequency:    10  MH2 
Temperature  of  Water  Bath:    18  "C 
Gate  Width:    10  us  T^:  1( 

Transducer  Position:      34  cm 
Sample  Position:    52.7  cm 
Beam  Width:   0.27  cm 
Calibration  Run: 

Reflector  Position 


Pulse  Repetition  Rate;    1  kHz 
Pulse  Width:    4  vs 


Reflector  Positi 


Beam  Cross -Sect ion :  0.057  cm^ 


Attenuator  Readi 


58 


Reference  on  RMS  Meter:     0.003  V,  -8  dB 
RCDistance  between  Transducer  and  Sample)  =  15.4 
a [Attenuation  Coefficient  for  Sample)  =  .1 
c(Velocity  of  Propagation)  =  1.58 

DATA 


1     11.5  2     10  3     11  4  9,5 

89  9      8  10  10  11  9.5 

IS  11 

REDUCED  DATA 

1     0.017  2     0.023      3    0.015      4  o.Oll 

8    0.01  9     0.008    10     0.012    H     0.011  1 
15  0.015 
Average : 


5  10.5 
12  15 


cm  -31.19 
0.29 

Nepers/cm  1,68 
-20.59 

cm/second  -29.19 


14  15 


SCATTERING  COEFFICIENT 
0.014  6  0.03  7  0.06 
0.038    13     0.06      14  0.038 


Fig.  1.    Sample  data  collection  sheet.    For  each 
specimen  at  each  frequency  the  data  shown 
were  recorded  by  the  experimenter. 

us  to  check  for  R"2  echo  dependence,  eq.  (1),  as 
proof  of  far-field  conditions,  and  relates  the  at- 
tenuator reading  at  the  surface  of  the  sample  to 
the  attenuator  reading  at  a  reference  position  with- 
in the  tank.    Readings  from  the  reference  position 
are  used  to  monitor  system  gain.    The  next  four 
lines  retain  data  which  apply  to  this  particular 
experiment.    The  distance,  R,  shown  was  calculated 
from  the  position  of  transducer  and  sample  carriers 
shown  at  the  top  of  the  page  and  the  use  of  correc- 
tions for  the  distance  between  the  active  surfaces 
of  transducer  and  sample  and  the  holder  position. 
Attenuation  coefficients  and  velocity  of  propaga- 
tion (inadvertently  recorded  as  cm/second  rather 
than  meters/second  used  in  calculation)  were  ob- 
tained from  the  literature.    The  numbers  shown  at 
the  right  side  of  the  figure  were  notes  made  during 
calculation  in  which  various  factors  in  eq.  (3) 
were  expressed  in  terms  of  dB.    Under  data  we  list 
the  attenuator  readings  necessary  to  keep  the  RMS 
meter  at  its  reference  reading. 

The  result  of  final  calculation  of  scattering 
coefficient,  nd,  is  shown  at  the  bottom  of  the 
page. 

3.    Results  and  Discussion 

The  experimentally  determined  values  of  scatter- 
ing coefficient  are  plotted  in  figures  2  and  3. 


0.03 


0.02 


0.01 


0.005  - 


0.001 


^  0.0005 


0.0001 


For  Myocardium 
O  Normal 
•  Fixed 


till 


1 


2       3     4    5  6 
Frequency  (MHz] 


10 


Fig.  2.    Measured  backscattering  coefficient  for 

the  myocardium  as  a  function  of  frequency. 
Experimental  points  are  the  average  of 
30  to  45  separate  measurements  on  five  or 
six  samples.    Bars  indicate  the  standard 
deviation  vs_.  frequency  slopes  for  Rayleigh, 
fourth  power  scattering  and  for  third  power 
scattering. 

One  point  using  tissue  fixed  with  10  percent  for- 
malin is  included  for  comparison.    The  data  show 
that  liver  is  a  much  stronger  scatterer  than  heart 
tissue  at  the  lower  MHz  frequencies.    Also,  the 
fixed  tissue  has  a  stronger  scattering  than  the 
normal  tissue  in  each  case.    The  backscattering 
from  both  myocardium  and  liver  increases  with  fre- 
quency.   Figure  2  shows  reference  lines  drawn 
through  arbitrary  points  to  give  the  attenuation 
slope  for  both  fourth  power  (Rayleigh)  and  third 
power  scattering.    At  the  higher  frequencies,  the 
heart  tissue  appears  to  be  approaching  approximate 
fourth  power  dependence  and  the  liver  tissue  less 
than  a  third  power  dependence.    It  must  be  recog- 
nized that  the  program  is  still  in  its  early  stages 
since  only  a  few  data  points  have  been  taken  and 
positive  conclusions,  particularly  with  respect  to 
frequency  dependence,  are  not  warranted.  Scatter- 
ing does  appear  to  increase  with  frequency,  how- 
ever. 

Since  the  scattering  at  10  MHz  is  still  increas- 
ing with  frequency  (fig.  2),  it  appears  that  the 


155 


0.03 


0.02 


0.01 


01 


S  0.005 


0.001 


For  Liver 
A  Normal 
▲  Fixed 


1 


1 


2  3        4  5 

Frequency  (MHz) 


Fig.  3, 


Backscattering  coefficient  for  liver  tis- 
sue as  a  function  of  frequency. 

structure  responsible  for  producing  the  scattering 
must  be  smaller  than  the  wavelength.    For  a  spheri- 
cal particle  it  is  known  that  the  circumference 
must  be  less  than  one  wavelength,  indicating  that 
the  diameter  of  such  tissue  elements  in  liver  is 
less  than  30  microns. 

From  these  data,  it  is  possible  to  estimate  the 
contribution  to  attenuation  caused  by  scattering 
alone.    In  cases  where  the  angular  dependence  is 
known  the  total  scattering  loss,  as,  can  be  found 
from 


4tt 


|i  ,o)da) 


fluence  the  total  attenuation.    Since  the  scatter- 
ing contribution  is  a  strong  function  of  frequency, 
we  might  expect  that  at  a  sufficiently  high  fre- 
quency the  attenuation  of  some  tissues  or  tumors 
might  increase  at  a  rate  greater  than  the  first 
power  dependence  found  by  previous  investigators 
working  at  lower  frequencies.    If  verified,  such 
behavior  could  have  great  importance  for  tissue 
characterization,  whether  it  be  done  using  scat- 
tered waves  or  measurements  of  attenuation. 

References 

[1]    Reid,  J.  M. ,  The  Scattering  of  Ultrasound 
by  Tissues,  in  Ultrasonic  Tissue  Charac- 
terization, M.  Linzer,  ed..  National  Bureau 
of  Standards  Spec.  Publ.  453,  pp.  29-47 
(U.S.  Government  Printing  Office,  Washing- 
ton, D.C.,  1976). 

[2]    Sigelmann,  R.  A.  and  Reid,  J.  M.,  The  analy- 
sis and  measurement  of  ultrasound  backscat- 
tering from  an  ensemble  of  scatterers  excit- 
ed by  sine  wave  bursts,  J.  Acoust.  Soc.  Am. 
35^,  1351  (1973). 

[3]    Shung,  K.  K.,  Sigelmann,  R.  A.,  and  Reid, 

J.  M. ,  The  scattering  of  ultrasound  by  blood, 
IEEE  Trans,  on  Biomedical  Engineering,  BME-23, 
6,  460  (1976). 


[4]    Shung,  K.  K.,  Sigelmann,  R.  A.,  and  Reid, 
J.  M. ,  The  Scattering  of  Ultrasound  by  Red 
8    10  Blood  Cells,  In  Ultrasonic  Tissue  Characteriza- 

tion, M.  Linzer,  ed..  National  Bureau  of 
Standards  Spec.  Publ.  453,  pp.  207-212  (U.S. 
Government  Printing  Office,  Washington,  D.C., 
1976). 

[5]    Nicholas,  D.,  The  Application  of  Acoustic 

Scattering  Parameters  to  the  Characterization 
of  Human  Soft  Tissues,  in  1976,  IEEE  Ultra- 
sonic Symposium  Proc,  pp.  64-69  (1976),  No. 
76CH  1120-5SU. 

[6]    Wells,  P.  N.  T.,  Physical  Principles  of  Ultra- 
sonic Diagnosis  (Academic  Press,  London  and 
New  York,  1969]". 

[7]    Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of  ab- 
sorption of  ultrasound  In  liver,  J.  Acoust. 
Soc.  Am.  50,  192  (1971). 


I    I   I  I 


(4) 


where 


Dd 
I 

0 


the  solid  angle 

differential  scattering  coefficient 
the  Incident  direction  of  the  wave 
the  direction  of  observation 


As  a  first  approximation,  let  us  assume  that  the 
scattering  from  liver  is  isotropic,  then. 


4Trn 


180 


At  10  MHz  frequency  the  attenuation  coefficient  for 
liver  due  to  scattering  alone  is  then  0.3  neper/cm. 
This  loss  is  about  28  percent  of  the  total  attenua- 
tion found  in  the  literature  [7j~. 

This  investigation  has  obtained  experimental 
evidence  that  the  scattering  from  within  functional- 
ly homogenous  tissue  may  be  strong  enough  to  in- 


[8]    Yuhas,  D.  E.,  Mimbs,  J.  W. ,  Miller,  J.  G., 
Weiss,  A.  N.,  and  Sobel ,  B.  E.,  Changes  in 
Ultrasonic  Attenuation  Indicative  of  Regional 
Myocardial  Infarction,  in  Ultrasonics  in 
Medicine,  D.  White,  ed.,  Vol.  3  (Plenum  Press, 
New  York,  1976). 

[9]    Gramiak,  R. ,  Hunter,  L.  P.,  Lee,  P.  P.  K. , 
Lerner,  P.  H.,  Schenk,  E.  ,  and  Waag,  K.  C, 
Diffraction  Characterization  of  Tissue  Using 
Ultrasound,  in  1976,  IEEE  Ultrasonics  Sympos- 
ium Proc. ,  pp.  60-63  (1976),  No.  76CH  1120-5SU. 


^It  has  to  be  noted  this  figure  obtained  was  based 
on  a  very  rough  assumption.    In  reality,  the 
scattering  from  liver  is  not  isotropic  as  indicated 
by  the  experimental  results  of  Gramiak  et_  aj_.  [9]. 
Our  measurements  do  average  n  over  many  directions, 
so  we  feel  the  conclusions  are  warranted. 


156 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,   Spec.  Publ.   525    (U.S.  Government  Printing  Office,  Washington,   D.C.,  1979). 


DEPENDENCE  OF  ULTRASOUND  BACKSCATTER  FROM  HUMAN  LIVER  TISSUE 
ON  FREQUENCY  AND  PROTEIN/LIPID  COMPOSITION 


M.  Freese 

Radionics  Ltd. 
Montreal,  Quebec,  Canada 

E.  A.  Lyons 

Winnipeg  Health  Sciences  Center 
Winnipeg,  Manitoba,  Canada 


The  dependence  of  ultrasonic  volume  backscatter  on  frequency,  lipid  and  protein  con- 
tent in  normal  and  fatty  human  (post-mortem)  liver  tissue  was  investigated  with  a  view 
towards  the  possible  development  of  a  noninvasive  quantitative  diagnostic  test  for  fat- 
ty liver.    Mean  values  of  the  backscatter  coefficient,  its  range  and  frequency  de- 
pendence were  determined  for  normal  liver  in  the  1  to  5  MHz  range  and  for  fatty  liver 
at  frequencies  of  2.25  and  3.56  MHz.    The  results  indicate  that  frequencies  below  about 
2  MHz  should  be  avoided  for  quantitative  measurements  of  the  backscatter  in  liver. 
Although  the  measurements  revealed  considerable  variation  in  the  backscatter  levels  of 
normal  liver,  the  backscatter  levels  in  moderately  and  severely  fatty  liver  were  sig- 
nificantly greater  than  the  normal  range.    Simple  linear  correlation  of  the  backscatter 
magnitude  with  the  lipid  content  for  21  samples  (10  normal,  8  fatty,  1  cirrhotic,  2 
other  abnormal)  yielded  a  value  of  0.94,  significant  at  the  1  percent  level. 


Key  words:    Backscatter  frequency  dependence;  cirrhosis;  composition-dependent 

scattering;  fatty  liver;  stochastic  scattering;  tissue  characterization; 
ultrasound  attenuation;  ultrasound  diagnosis;  ultrasonic  tissue 
scattering. 


1.  Introduction 

Quantitative  data  on  the  dependence  of  ultra- 
sound scattering  on  the  physical  structure  and 
composition  of  tissue  is  needed  to  understand  the 
underlying  scattering  processes,  to  aid  the  de- 
velopment of  optimum  diagnostic  scattering 
methods  and  to  serve  as  a  reference  in  clinical 
applications.    From  the  standpoint  of  these  ob- 
jectives, specifically,  the  possible  development 
of  a  noninvasive  diagnostic  test  for  fatty  liver 
[1]^,  this  paper  describes  the  results  of  a 
series  of  quantitative  measurements  of  ultrasonic 
volume  backscatter  in  the  1  to  5  MHz  range  for 
normal  and  fatty  post-mortem  human  liver  as  a 
function  of  frequency  and  protein-1 ipid  content. 
The  results  suggest  that  a  noninvasive  quantita- 
tive test  for  fatty  liver  based  on  the  magnitude 
of  the  backscatter  is  feasible  if  the  in  vivo  mea- 
surement  problems  can  be  overcome. 

2.  Methodology 

The  pulse  echo  technique  and  the  calibration 
procedure  employed  for  the  backscatter  measurements 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


have  been  described  in  detail  elsewhere  [2].  Es- 
sentially, the  received  backscatter  from  the  tissue 
is  range  gated,  passed  through  either  a  square-law 
or  linear  envelope  detector  and  integrated;  time 
varied  gain  is  employed  to  compensate  for  sample 
attenuation.    The  measurements  are  referenced  to 
stainless  steel  ball  bearing  targets  of  known  cross- 
section. 

A.    Backscatter  Parameters 

The  parameters  used  to  describe  the  backscatter- 
ing  are  the  average  backscatter  coefficient  <n> 
and  the  average  "envelope"  backscatter  coefficient 
<r>,  the  brackets  denoting  ensemble  averaging.  For 
monochromatic  (CW)  frequencies,  <fi>  is  identical  to 
the  average  backscatter  cross-section  per  unit 
volume,  while  for  pulses,  the  difference  between 
<fi>  and  the  corresponding  CW  value  is  dependent  on 
the  absorption  coef f icient--path  length--pulse 
bandwidth  product  and  the  scatterer  frequency 
responses  [2,3].  (The  maximum  estimated  error  due 
to  the  finite  pulse  bandwidth  was  approximately 
0.1  dB  in  these  measurements.)    The  envelope  back- 
scatter  coefficient  <r>  is  proportional  to  the 
average  magnitude  of  the  backscatter  signal  but  is 
normalized  to  be  consistent  with  the  statistical 
relation  <fi>  =  <v'^>. 


157 


B.    Signal  Processing 

Stochastic  signal  process  theoretical  results 
were  employed  to  determine  appropriate  thresholds 
for  refining  of  the  raw  data.    Scattering  by  the 
larger  blood  vessels  in  the  liver  poses  a  problem 
because  its  contribution  tends  to  be  largely  extra- 
neous to  the  scattering  process  of  interest.  To 
reduce  this  interference,  the  raw  data  were  refined 
by  rejecting  individual  measurements  with  values  of 
the  coefficient  of  variation  y  >  0.81,  the  coeffi- 
cientjOf  variation  being  defined  as  y  =  {<r'^>  - 
<r>2)^/<r>.    The  presence  of  isolated  large  specu- 
lar echoes  tends  to  increase  the  expectation  value 
of  Y  (=  0.52  for  a  random  process  described  by  a 
Rayleigh  distribution  [4]).    Ordinarily,  the  proba- 
bility of  Y  exceeding  0.81  is  about  7  percent.  The 
A-scan  time  averaging  employed  reduced  this  proba- 
bility to  5  percent  at  1.1  MHz  decreasing  to  0.5 
percent  at  3.56  MHz. 

To  further  improve  the  averaging  and  avoid  the 
long  delay  path  required  for  far-field  measure- 
ments, the  measurements  were  performed  in  the  near 
fields  of  the  transducers.    The  measured  values 
were  then  converted  to  equivalent  far-field  values 
using  the  experimental  results  obtained  by  Freese 
[5];  the  same  transducers  were  used  in  the  present 
measurements^.    The  near-  to  far-field  conversion 
factors  ranged  from  +  1.4  and  +  1.1  dB  at  1.1  MHz 
to  -  1.1  dB  at  4.88  MHz. 

Time  varied  gain  based  on  running  mean  absorp- 
tion estimates  were  employed  during  the  measure- 
ments with  appropriate  corrections  based  on  the  mea- 
sured absorptions  being  applied  at  the  completion 
of  the  experiments  [6]. 

C.  Measurements 

The  ultrasonic  measurements  on  the  liver  samples 
were  conducted  with  two  exceptions  within  one  to 
three  days  post-mortem.    The  samples  were  removed 
from  the  cadaver  prior  to  any  postmortem  infusion 
procedures,  and  were  cut  from  the  anterior  portion 
of  the  liver  extending  below  the  ribs,  with  the 
backplane  of  the  sample  sliced  roughly  parallel  to 
the  external  liver  surface.    With  the  samples  im- 
mersed in  physiological  saline  held  at  20  °C  and 
the  ultrasonic  beam  incident  perpendicularly  on  the 
external  liver  surface,  the  backscatter  was  measur- 
ed at  two  depths  extending  from  0.6  cm  to  1.8  cm, 
and  from  1.2  cm  to  2.4  cm  below  the  surface,  re- 
spectively.   Four  separate  measurements  were  made 
at  each  depth.    The  total  insonified  volume  averag- 
ed about  10  cm^  per  sample  (5  cm^  per  sample  in  the 
case  of  the  1.1  MHz  measurements). 

The  absorptions  were  measured  by  means  of  a  sub- 
stitution technique.    To  the  extent  possible,  all 
of  the  measurements  were  carried  out  under  double- 
blind  conditions. 

D.    Diagnostics  and  Biochemical  Assays 

Subsequent  to  the  ultrasonic  measurements,  the 
samples  (nineteen  male,  six  female)  were  classified 

2The  conversion  factors  (as  a  function  of  trans- 
ducer distance)  were  determined  from  scattering 
measurements  on  model  media  containing  random 
scatterers  whose  dimensions  were  comparable  to 
tissue  cells.    Corresponding  data  obtained  for 
whitefish  muscle  tissue  were  generally  found  to 
be  within  0.5  dB  of  the  model  media  values. 


on  the  basis  of  clinical  and  autopsy  data  as  either 
normal  or  abnormal.    The  abnormal  group  (eight  male, 
four  female)  was  then  further  categorized  accord- 
ing to  the  lipid  content--fatty  (seven  male,  two 
female)  or  nonfatty.    One  normal  and  one  fatty 
sample  were  rejected  because  of  technical  problems. 
The  lipid,  protein,  and  moisture  contents  of  the 
samples  were  determined  by  B.  Guy  Hunt  Laboratories 
(Winnipeg)  using  standard  chemical  analytical 
methods . 

3.    Results  and  Discussion 

The  average  lipid  content  and  physical  data  for 
the  normal  samples  (table  1)  are  in  good  agreement 
with  generally  accepted  values  [7];  however,  the 
protein  content  appears  to  be  about  20  percent  low. 
Detailed  data  on  the  abnormal  samples  is  given  in 
table  2. 

Table  1.    Composition  and  physical  characteristics 
of  normal  liver  samples. ^ 


Factor  Mean  Standard 

deviation 


Total  lipid  {%)^                       4.8  1.1 

Protein  {%  nitrogen  x  6.25)  14.6  2.1 

Moisture  {%)  77.5  2.8 

Liver  weight  (g)  1680  280 

Age  of  deceased  (y)                    55  26 

Weight  of  deceased  (kg)             67  9 


^Eleven  male,  one  female,  ranging  in  age  from 

,15  to  93  years. 
Estimated  accurace  of  liqid  analytical  deter- 
mination ±  (0.1  {%  lipid)  +  ]%). 


A.    Normal  and  Fatty  Liver  Tissue 
Ultrasound  Attenuation 

The  attenuation  was  measured  in  eight  of  the 
normal  samples,  and  yielded  a  mean  value  of  0.64 
dB/cm/MHz  for  the  frequency  normalized  absorption 
coefficient,  a/f.    Possibly  reflecting  a  lower 
protein  content,  this  value  is  about  10  percent 
lower  than  the  mean  value  quoted  by  Wells  [8]  for 
the  3  to  5  MHz  range.    The  average  attenuation  for 
the  normal  samples  at  the  2.25,  3.56  and  4.88  MHz 
frequencies  were  1.5,  2.3  and  3.2  dB/cm,  respectively. 

In  the  case  of  the  fatty  livers,  the  absorptions 
were  determined  for  all  but  the  7.7  percent  lipid 
sample  (table  2).    Excluding  the  45.6  percent  lipid 
sample,  which  exhibited  anomalously  high  absorption, 
the  average  value  of  a/f  for  the  fatty  livers  was 
0.86  dB/cm/MHz  or  about  one  third  greater  than 
for  the  normal  livers. 

The  two  extremely  fatty  samples  (table  2)  were 
rather  interesting  for,  although  similar  in  ap- 
pearance, they  had  markedly  different  absorptions. 
The  35.5  percent  lipid  sample  measured  2.74  dB/cm 
at  3.56  MHz  (six  replicates)  which  is  only  slight- 
ly more  than  in  normal  liver,  while  the  45.6  per- 
cent lipid  sample  measured  a  remarkable  14  dB/cm 
(three  replicates).    The  reason  for  this  anomalous- 
ly high  absorption  is  not  certain  but  it  seems  to 
point  to  the  possible  presence  of  gas  bubbles  in 
the  latter  sample,  despite  the  fact  that  the  liver 
sample  was  obtained  one  day  post-mortem.    Also,  un- 


158 


Table  2.    Composition  and  clinical  data  for  the  abnormal  samples. 


Group        Total    Protein    Moisture               Clinical  information  Autopsy  results 
lipid                              sex    age    height   weight        history        Liver  Liver 
 %  %  %  y        m  kg  weight,  g  appearance 


Fatty 


7.7 
9.0 


11.4 
14.7 


14.9 
19.1 


35.5 
45.6 


16.0 
14.0 


11.4a 
13.2 


13.2 
11.9 


10.6 
8.4 


72.2 
71.1 


74.3 
64.1 


71.2 
70.1 


52.5 
44.4 


M  — 
M  26 


M  — 
M  45 


M  54 
F  69 


M  60 
M  50 


1.63 
1.70 


1.65 
1.80 


1.63 


1.82 
1.68 


52 
69 


64 

59 


66 


84 
117 


alcohol  ism 

(blood  alcohol 
150  mg) 

alcohol  ism 


(blood  alcohol 
16  mg) 

alcohol  ism 


1630 
1690 


1940 
1900 


pale  yellowish 
alcoholic  hepa- 
titis, mod.  fat 
infilt. 


1600  pale 


3450 
5600 


yell ow 

yellow,  fatty 


Nonfatty 


3.6  18.1 


3.7  15.2 

3.8  16.8 


79.8 


79.6 

78 


M  64 


91 
76 


1.74 


1.56 
1.51 


82  cirrhosis 


1730 


40  diabetes  (diet)  1220 
37  —  730 


yellow,  firm 
nodul ar 

adv.  cirrhosis 
smal  1 


Estimate. 


like  the  35.5  percent  lipid  sample,  the  45.6  per- 
cent sample  was  buoyant  in  the  0.1  N  saline  solu- 
tion. 

B.    Normal  Liver  Backscatter  Coefficient, 
Range  and  Frequency  Dependence 

The  backscatter  coefficients  and  corresponding 
envelope  backscatter  coefficients  for  the  normal 
samples  are  plotted  as  a  function  of  frequency  in 
figure  1(a)  and  (b).    (A  number  of  single  data 
points  are  also  included.)    The  ranges  of  the  back- 
scatter  coefficients  at  each  of  the  measurement 
frequencies  are  extremely  broad.    For  example,  at 
2.25  mHz,  <^>  varies  from  0.17  to  2.9  mm2/cm3— a 
range  of  over  12  dB.    With  the  exception  of  a 
single  low  value  at  4.88  MHz,  which  may  have  been 
the  result  of  a  faulty  measurement,  the  largest 
range  is  exhibited  by  the  1.1  MHz  values.    It  is 
tempting  to  attribute  this  primarily  to  first- 
order  diffraction  effects  [9,10]  and  to  the  great- 
er statistical  fluctuation  at  this  frequency,  but 
a  more  careful  analysis  of  the  measurements  does 
not  seem  to  support  this  view.    While  large  fluc- 
tuations in  the  individual  measurement  values  were 
observed  in  many  of  the  samples,  when  the  measure- 
ments were  repeated  on  the  same  sample  using  slight- 
ly different  aspect  angles  and  insonified  sites, 
the  mean  values  seldom  differed  by  more  than  1  or 
2  dB.    Furthermore,  if  we  take  into  account  the 
dimensions  of  the  blood  vessels  in  terms  of  wave- 
length, the  decrease  of  the  highest  <fi>  values  be- 
tween 1.09  and  2.25  MHz  is  consistent  with  oblique 
incidence  scattering  by  cylinders  of  corresponding 
radii  in  the  transition  region  from  Rayleigh  to 
geometrical  scattering  (i.e.,    the  vessels  will 
tend  to  scatter  specularly  with  increasing  fre- 


quency).   Since  the  larger  blood  vessels  tend  to 
radiate  towards  the  liver  surface,  the  oblique 
aspect  angles  would  be  more  probable  in  our  measure- 
ments resulting  in  an  effective  decrease  in  the 
backscatter  at  the  higher  frequencies.    In  contrast 
to  the  vessel  orientations,  the  orientations  of  the 
liver  lobule  "fascia"  appear  essentially  random 
over  the  extent  of  the  insonified  volumes.  This 
would  lead  one  to  expect  the  ensemble  averaged 
backscatter  to  be  independent  of  aspect  angle.  In 
attempting  to  answer  the  question  concerning  the 
relative  contributions  of  the  liver  lobules,  blood 
vessels  and  other  liver  structures  to  the  overall 
scattering  process  [11],  our  results  would  seem  to 
indicate  that  scattering  by  the  larger  vessel  in- 
homogeneities,  if  present  in  the  insonified  volume, 
will  tend  to  dominate  the  backscatter  below  about 
1.5  MHz. 

The  average  frequency  dependence  of  <fi>  over  the 
2.25  to  3.56  MHz  range  is  given  by  f0-8(fig.  1(a)), 
decreasing  to  about  f°  in  the  1.09  to  2.25  MHz  in- 
terval, and  increasing  to  f^-^  between  3.56  and 
4.88  MHz  (if  we  exclude  the  4.88  MHz  value  for  the 
borderline  sample).    While  there  are  too  few  values 
to  provide  more  than  a  rough  estimate  of  the  fre- 
quency dependence  in  the  latter  two  intervals, 
these  values  are  in  good  agreement  with  those  ob- 
tained by  Nicholas  [12]  using  spectral  analysis. 
The  relatively  low  frequency  dependence  observed 
illustrates  the  effect  that  the  presence  of  even  a 
relatively  small  number  of  "geometric!  region" 
scatterers  can  have  on  the  composite  frequency  de- 
pendence.   For  this  reason  the  frequency  dependence 
of  diffuse  volume  backscatter  may  not  be  as  aseful 
a  parameter  for  some  tissue  diagnostic  applications 
as,  the  magnitude,  the  angular  dependence  or  the 
absorption. 


159 


10 


2J3  3D  5D 
Frequency  (MHz) 

(a) 


70 


2.0  3JD 
Frequency  (MHz) 

(b) 


50 


7.0 


Fig.  1.    (a)  Backscatter  coefficients  as  a  function  of  frequency  for  samples  of  normal 
liver  tissue.    The  logarithmic  mean  values  of  <n>  for  the  2.25  and  3.56  MHz 
data  are  indicated  by  connecting  heavy  barred  lines.    The  dashed  lines  corre- 
spond to  ±  6  dB  about  the  logarithmic  means  and  bound  essentially  all  of  the 
measured  values  in  this  frequency  range.    The  sample  with  the  lowest  back- 
scatter  coefficients  suggests  a  borderline  sample.    Its  moisture  content  of 
81  percent  was  second  highest  of  the  normal  samples, 
(b)  Corresponding  envel ope  backscatter  coefficients. 


Similar  comments  apply  to  the  envelope  backscat- 
ter coefficient  shovm  in  figure  1(b)  although  the 
frequency  dependence  is  only  half  as  great  (or  less 
if  the  scattering  process  is  inhomogeneous) .  The 
principal  advantages  of  using  <r>  are  that  it  sim- 
plifies the  electronics  requirements  and  is  less 
sensitive  to  isolated  large  extraneous  echoes 
thereby  reducing  the  consequent  error.    For  these 
reasons,  we  will  consider  mostly  <r>  instead  of 
<n>  in  the  following  paragraphs.    However,  to  ob- 
tain an  estimate  of  <n>  one  need  only  compute 
(1  +  Y^)<r>2  using  the  average  measured  value  of 
0.6  for  Y  instead  of  the  theoretical  value  of  0.52. 

C.    Backscatter  Dependence  on  Lipid  and 
Protein  Content 

Values  of  <r>  were  compared  with  the  lipid  and 
protein  contents  of  both  the  normal  and  abnormal 
samples.    Figure  2  shows  the  2.25  MHz  values  of 
<r>  as  a  function  of  the  protein  content  (P)  for 
normal  samples.    The  value  of  the  simple  linear 
correlation  coefficient  rj-p  =  0.776  and  is  signifi- 
cant at  the  1  percent  level  (t-test  assuming  rpp  = 
0;  ten  samples).    Comparable  results  were  obtained 
at  the  higher  frequencies.    This  implies  that  in 
normal  liver  inherent  differences  in  the  protein 
content  may  result  in  <r>  varying  as  much  as  9  dB. 


(Conversely,  taking  the  protein  into  account  and 
assuming  the  scattering  processes  add  in  quadrature, 
the  12  dB  range  of  the  2.25  MHz  values  in  figure 
1(a)  and  (b)  can,  in  principle,  be  reduced  by 
4.5  dB.) 


2.0 


0.0 


Fig.  2 


fr  =  2.25  MHz 


14 

Protein 


16 

(7o  nitrogen 


20 


6.25) 


Dependence  of  <r>  on  protein  content  in 
normal  liver;  the  correlation  coefficient 
r^p  =  0.776. 


160 


No  significant  correlation  was  observed  at  1.1 
MHz.    This  fact  coupled  with  the  previous  observa- 
tions would  seem  to  indicate  that  frequencies  below 
about  2  MHz  should  be  avoided  for  quantitative  mea- 
surements of  volume  backscatter  from  liver  tissue. 

In  contrast  to  the  correlation  of  the  backscat- 
ter with  protein  content,  the  correlation  of  <r> 
with  the  lipid  content  of  the  normal  samples  was 
not  significant.    The  presence  of  glycogen  and 
other  minor  hepatic  constituents  (moisture  tends 
to  correlate  inversely  with  lipid  and  protein)  has 
been  neglected  but  their  contribution  is  not  likely 
to  be  significant  in  normal  liver. 

The  effect  of  the  lipid  becomes  apparent  in  the 
fatty  liver  samples,  which  ranged  from  7.7  to  45.6 
percent  lipid.    The  backscatter  and  lipid  content 
(L)  are  highly  correlated  (rp^  =  0.980  and  0.978 
at  2.25  and  3.56  MHz,  respectively)  and  suggest  a 
roughly  linear  relationship  between  the  two  as 
shown  in  figure  3  for  the  2.25  MHz  values. 

However,  in  view  of  the  limited  number  of  mea- 
surements and  the  possibility  that  the  45.6  percent 
lipid  sample  may  have  contained  gas  bubbles,  the 
latter  is  probably  at  least  partly  due  to  coinci- 
dence since  a  linear  dependence  for  a  lipid  content 
aoproaching  50  percent  would  seem  unlikely. 

The  histological  appearance  of  mildly  and  severe- 
ly fatty  livers  are  shown  in  figure  4(a)  and  (b). 
It  can  be  seen  from  these  pictures  that  the  struc- 
tural changes  that  occur  in  hepatic  tissue  with  in- 
creasing lipid  content  span  the  spectrum  from  nearly 
normal  to  grossly  abnormal.    If  we  combine  the  two 


Fig.  4.    (a)  Mildly  fatty  liver  in  an  alcoholic.    Many  of  the  hepatocytes  contain  large  lipid  globules 
(clear  areas  in  the  stained  cytoplasm).    Magnification  25X. 

(b)  Severe  fatty  liver  in  an  alcoholic.  The  nuclei  of  the  cells  are  displaced  by  large  lipid 
globules;  little  cytoplasm  remains.    Magnification  25X. 


5,0 


4.0 


3.0 


20 


1,0 


0.0 


20  30 
Total   Lipid  (  %  ) 


40 


50 


Fig.  3 


Dependence  of  <r>  on  lipid  content  in  fatty 
liver;    the  correlation  coefficient  rpL  = 
0.980;  ff  =  2.25  MHz. 


groups  of  samples  and  take  the  protein  content  into 
account,  the  multiple  linear  regression  equation 
for  <r>  at  2.25  MHz  becomes  <r>  =  -1.4  +  O.IOL  + 
0.12P  with  rp  Lp  =  0.966.    This  indicates  that  the 
effective  scattering  contributions  by  the  protein 
and  lipid  (either  directly  or  indirectly)  are  rough- 
ly comparable  on  a  percentage  weight  basis,  with 
the  protein  contributing  about  40  percent  more  to 
<fi>  at  this  frequency  than  the  lipid  for  equal  per- 


161 


cent.ages  of  the  two  constituents.    The  correspond- 
ing result  at  3.56  MHz  was  <r>  =  -1.9  +  O.IOL  + 
0.17P  with  rp  |_p  =  0.939.    In  this  case,  data  was 
available  for'only  eight  normal  and  six  fatty 
samples.    Additional  measurements  will  be  needed  to 
establish  if  the  difference  in  the  relative  contri- 
butions implied  by  these  two  regressions  (and  there- 
fore the  frequency  dependence  of  the  normal  and  fat- 
ty samples)  is  signif icaitit. 

For  predictive  purposes  we  require  the  percent 
lipid  as  a  function  of  <r>  and  the  percent  protein. 
Moreover,  the  remaining  abnormal  nonfatty  samples, 
listed  in  table  2,  should  also  be  included  in  the 
regression.    The  resultant  linear  regression  equa- 
tion obtained  at  2.25  MHz  was  L  =  15.1  +  9.2<r> 
-1.2P,  with  r|_  j,p  =  0,971.    This  regression  with 
18  degrees  of  freedom  (21  samples)  and  F-test  value 
=  157  is  significant  at  the  1  percent  level.  Com- 
paring it  to  the  previous  2.25  MHz  regression,  the 
effect  of  the  additional  samples  is  seen  to  be 
minimal . 

5  0  (- 


0  10  20  30  40  50 

Total  Lipid  (  %  ) 

Fig.  5    Dependence  of  <r>  on  the  lipid  content 
for  the  combined  group  of  normal  and  ab- 
normal samples;  the  correlation  coeffi- 
cient rj,^  =  0.943. 

In  potential  diagnostic  applications  the  percent 
protein  would  most  likely  not  be  available.    The  re- 
sultant simple  two  variable  linear  regression  equa- 
tion obtained,  L  =  -442  +  11.0<r>,  is  graphed  in 
figure  5.    The  correlation  coefficient        =  0.943 
is  essentially  the  same  as  when  the  nonfatty  ab- 
normal samples  were  excluded.    Figure  5  shows  that 
the  backscatter  coefficients  are  substantially 
greater  for  the  moderately  and  extremely  fatty 
livers.    The  corresponding  regression  at  3.56  MHz, 
based  on  17  samples,  resulted  in  L  =  -6.2  +  11.3 
<r>  and  rL^  =  0.914.    In  view  of  the  limited  number 
of  samples  and  the  large  weighting  by  the  two  ex- 
tremely fatty  livers,  these  regressions  can  only  be 
considered  preliminary.    Nevertheless,  they  seem  to 
suggest  that  a  simple  quantitative  diagnostic  test 
based  on  the  magnitude  of  the  backscatter  is  feasi- 
ble in  principle. 


4.  Summary 

Experimental  values  of  the  volume  backscatter 
coefficients  and  estimates  of  their  distribution 
and  frequency  dependence  were  obtained  for  normal 
liver.    These  were  discussed  and  compared  from  the 
point  of  view  of  the  underlying  scattering  proces- 
ses, the  liver  composition  and  potential  diagnostic 
requirements.    The  results  indicate  that  in  the 
case  of  liver,  frequencies  under  2  MHz  should  be 
avoided  for  quantitative  diagnostic  backscatter 
measurements. 

Significant  correlations  of  the  backscatter  with 
protein  content  were  observed  in  normal  liver  and 
with  both  protein  and  lipid  content  in  abnormal 
fatty  liver.    In  the  case  of  the  latter,  regressions 
significant  at  the  1  percent  level  were  obtained. 

Acknowledgment 

This  work  was  supported  in  part  by  the  Depart- 
ment of  Environment,  Freshwater  Institute,  Winnipeg, 
Canada . 

References 

[1]    Wells,  P.  N.  T.,  McCarthy,  C.  F.,  Ross,  F.  M. 
G.,  and  Read,  A.  E.  A.,  Comparison  of  A-scan 
and  compound  B-scan  ultrasonography  in  the 
diagnosis  of  liver  disease,  Br.  J.  Radiol.  42, 
818-823  (1969). 

[2]    Freese,  M.  and  Hamid,  M.  A.  K. ,  Lipid  Content 
Determination  in  Whole  Fish  Using  Ultrasonic 
Pulse  Backscatter,  1974  Ultrasonics  Symp. 
Proc,  IEEE  Cat.  #74  CHO-896  15U,  69-76  (1974). 

[3]    Freese,  M. ,  Quantitative  diagnostic  measure- 
ments of  ultrasound  tissue  scattering  (in 
preparation). 

[4]    Ol'shevskii,  V.  V.,  Characteristics  of  Sea 
Reverberation  (Consultants  Bureau,  New  York, 
1967). 

[5]    Freese,  M. ,  Comparison  of  Near-  and  Far- 
Field  Measurements  of  Diffuse  Ultrasonic  Tis- 
sue Volume  Backscatter,  1975  Ultrasonics 
Symp.  Proc.  IEEE  Cat.  #75  CHO-994  45U,  33- 
36  (1975). 

[6]    Freese,  M.  and  Lyons,  E.  A.,  Ultrasonic  back- 
scatter  from  human  liver  tissue;  its  de- 
pendence on  frequency  and  protein/1 ipid  com- 
position, J.  Clin.  Ultrasound  (in  press). 

[7]    Schiff,  L.,  ed..  Diseases  of  the  Liver,  3rd 
ed.  (Lippencott  Co. ,  New  York,  1969). 

[8]    Wells,  P.  N.  T.,  Physical  Principles  of 

Ultrasonic  Diagnosis  (Academic  Press,  London 
and  New  York,  1969). 

[9]    Nicholas,  D.  and  Hill,  C.  R.,  Acoustic  Bragg 
diffraction  from  human  tissues.  Nature  257, 
305  (1975). 


162 


[10]    Lele,  P.  P.,  Mansfield,  A.  B.,  Murphy,  A.  I., 
Namery,  J.,  and  Senapati,  N.,  Tissue  Charac- 
terization by  Ultrasonic  Frequency-Dependent 
Attenuation  and  Scattering,  in  Ultrasonic 
Tissue  Characterization,  M.  Linzer,  ed.. 
National  Bureau  of  Standards,  Spec.  Publ. 
453,  pp.  167-196  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

[11]    Gramiak,  R.,  Hunter,  L.  P.,  Lee,  P.  P.  K., 
Lerner,  R.  M. ,  Schenk,  E.,  and  Waag,  R.  C, 
Diffraction  Characterization  of  Tissue  Using 
Ultrasound,  1976  Ultrasonics  Symp.  Proc, 
IEEE  Cat.  #76  CH1120-5SU,  60-63  (1976). 

[12]    Nicholas,  D.,  The  Application  of  Acoustic 
Scattering  Parameters  to  the  Characteriza- 
tion of  Human  Soft  Tissues,  1976  Ultrasonics 
Symp.  Proc,  IEEE  Cat.  #76  CH1120-5SU,  64- 
69  (1976). 


163 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ULTRASOUND  BACKSCATTERING  FROM  BLOOD:     HEMATOCRIT  AND 
ERYTHROCYTE  AGGREGATION  DEPENDENCE 


M.  Hanss  and  M.  Boynard 

Laboratoire  de  Biophysique 
UER  Exp^rimentale  de  M^decine  et  de  Biologie 
74,  rue  Marcel  Cachin  -  93000  Bobigny 

and 

UER  Biom^dicale  des  Saints-Peres 
45,  rue  des  Saints-Peres  -  75006  Paris 

The  ultrasound  back-scattering  of  blood  has  been  shown  to  be  increased  when 
the  erythrocytes  sedimentation  rate  is  high.    Moreover,  in  this  case,  temporal 
fluctuations  of  the  back-scattered  intensity  have  been  also  demonstrated.  To 
explain  these  results,  a  simplified  theory  of  ultrasound  back-scattering  by 
blood  is  presented.    The  model  of  the  blood  structure  used  is:    the  erythrocytes 
are  associated  to  form  spherical  transient  aggregates  having  the  same  cell 
number,  m.    The  acoustic  wave  is  scattered  by  these  clusters  called  "scattering 
units."    The  scattered  intensity  is  calculated  for  one  scattering  unit  taking  a 
spherical  elastic  model,  then  for  a  blood  unit  volume,  when  the  hemotocrit  H  is 
very  low.    It  is  found  that  the  back-scattered  intensity  is  proportional  to  H 
and  to  m.    When  H  is  not  negligible,  a  simplified  statistical  theory  is  proposed 
The  main  result  is  that  the  back-scattered  intensity  is  still  proportional  to  m 
but  that  it  passes  through  a  maximum  value  when  H  get  near  0.3.    An  explicit 
relation  between  the  mean  relative  value  of  the  intensity  fluctuation,  H  and  m 
is  given.    Therefore,  the  aggregation  state  of  blood  can  in  principle  be 
determined  through  ultrasound  scattering  studies.    Further  theoretical  develop- 
ments are  in  progress  so  as  to  take  into  account  the  distribution  of  the  number 
of  erythrocytes  in  each  kinetic  units  and  the  variation  of  this  number  with  H. 

Key  words:    Blood  erythrocytes  aggregation;  blood  hemotocrit;  ultrasound  back- 
scattering  from  blood. 


When  an  ultrasound  echographic  probe  is  placed 
on  top  of  a  vertical  cylindrical  tube  filled  with 
blood,  it  has  been  found  that  the  echo  pattern  of 
the  sedimenting  blood  varies  according  to  its 
sedimentation  rate  [1]^.  For  high  sedimentation 
rate  samples,  the  mean  back-scattered  intensity 
is  larger;  furthermore  the  echo  amplitude  at  a 
given  depth  in  the  erythrocyte  column  presents 
characteristic  temporal  fluctuations.    We  wish  to 
present  a  simple  theoretical  approach  in  order  to 
explain  these  phenomena.    Though  the  scattering 
of  ultrasound  by  blood  is  increasingly  used  in 
clinical  medicine,  the  fundamental  data  on  this 
property  are  very  scarce,  and  mainly  due  to  Reid 
and  co-workers  [2-5].    Fluctuations  (temporal  and 
spatial)  in  the  echo  amplitude  have  also  been  re- 
ported by  Atkinson  and  Berry  [6].    However,  none 
of  these  studies  can  give  a  straightforward  expla- 
nation of  our  experimental  results. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


Blood  Model 

The  red  blood  cell  (RBC)  concentration  is 
usually  defined  by  the  hematocrit  H,  that  is,  the 
time  average  of  the  RBC  volume  fraction  in  a  given 
blood  volume.    For  a  normal  blood  sample,  H  is 
rather  large  (H  =  0.45)  so  that  the  RBC  interact 
each  with  the  other.    We  will  describe  these  in- 
teractions by  assuming  that  m  RBC  are  mechanical- 
ly correlated  during  a  period  tc    larger  than  the 
ultrasound  pulse,  giving  rise  to  a  transient  ag- 
gregate which  we  assume  to  be  spherical,  with 
radius  A;  this  dimension  could  be  thought  of  as  a 
correlation  length.    We  will  also  suppose  that 
each  aggregate  has  the  same  number  m  of  RBC.  The 
ultrasound  wave  is  scattered  by  these  aggregates 
which  therefore  will  be  called  the  scattering 
units  (S.U.). 

If  V  is  the  number  of  S.U.  per  unit  volume  and 
V|^  the  RBC  volume,  the  usual  hematocrit  number  H 
is  given  by: 

H  =  mv,\)  n) 


165 


However  we  are  interested  in  the  volume  fraction 
of  the  S.U.  and  not  the  static  hematocrit  H.  We 
will  introduce  a  "dynamic"  hematocrit  number  H' 
as  follows: 

H'  =  vv' 
su 

where  viy  is  the  excluded  volume  of  the  S.U. 
(fig.  1)  which  takes  into  account  the  S.U.  volume, 
vsu.  and  the  trapped  plasma  between  the  S.U.: 


da  =  —  dfi 
i 


In  back-scattering  experiments,  the  angles  6 
and  (j)  values  are  180°  and  90°  respectively  (fig. 
2).    The  detecting  solid  angle  depends  on  the 
probe  surface  S  and  the  distance  x  between  the 
probe  and  the  S.U.    One  can  take  into  account 
these  parameters  by  introducing  a  geometrical 
factor  g(x)  =  S/x^. 


su 


su 


where  p^^  is  a  dimensionless  number  greater  than 
or  equal  to  1.    The  volume  of  one  S.U.  must  also 
take  into  account  the  trapped  plasma  inside  the 
S.U.: 


su 


"^^hPh 


where  p^  is  another  dimensionless  quantity  great- 
er than  or  equal  to  1.    If  the  RBC  were  perfectly 
packed  without  free  volume  for  the  plasma,  pj^ 
and  ph  would  be  equal  to  1.    For  an  imperfect 
packing,  they  are  greater  than  1.    The  dynamic 
hematocrit  H'  is  greater  than  H  because  in  the 
course  of  their  motions  the  RBC's  and  S.U.'s  car- 
ry with  them  a  fraction  of  the  plasma.    Its  value 
is: 


H' 


'hPhPh 


Using  this  model,  the  S.U.  radius  is  given  by: 


A  = 


4^  "^^Ph 


ll3 


(2) 


(3) 


Trapped  plasma 
Inside  the  S.U. 


one  S.U. 


'  Trapped 
,  ' '  plasma 
between 
^the  S.U. 


Fig.  1.    Blood  model  used,  showing  the  transient 
spherical  aggregation  (scattering  unit 
S.U.). 


Ultrasound  Back-Scattering  by  One  S.U. 

Let  io  be  the  incident  wave  intensity  per  unit 
area  of  a  spherical  scattering  center,  with  radi- 
us A  smaller  than  the  wavelength.    The  scattered 
intensity  per  unit  solid  angle  in  the  direction  e 
is  i[).    The  detecting  solid  angle  being  du,  the 
differential  scattering  cross-section  is: 


Emetting  probe 


Receiving  probe 
\ 


Fig.  2.    Scattering  coordinates. 

We  can  only  measure  intensities  emitted  and 
received  by  the  probe,  Iq  and  Iq.    They  differ 
from  io  and  ip  because  of  the  attenuation  of  the 
incident  and  scattered  beams  by  the  medium  be- 
tween the  probe  and  the  S.U.    They  also  differ 
because  the  ultrasonic  field  varies  with  distance. 
We  can  lump  all  these  factors  in  a  general  func- 
tion of  the  distance  x,  G(x)  so  that  one  can 
write 


(4) 


The  factor  G(x)  is  not  essential  in  our  final 
results.    It  could  be  determined  by  a  substitu- 
tion method  as  described,  for  instance,  by  Sigel- 
mann  and  Reid  [3]. 

The  differential  scattering  cross-section  is 
known  for  some  simple  models  [7].    Using  the  re- 
sults for  an  elastic  sphere,  the  back-scattered 
intensity  is  finally: 


I,G(x) 


w'^A^ 

9c^ 

V  C2p 

+  3 


2p  +  P. 


(5) 


where  c  and  Cg  are  the  sound  wave  velocities  in 
the  S.U.  and  the  plasma  respectively,  and  p  and 
Pp  the  corresponding  densities.    We  assume  that 
the  S.U.  coefficients,  c  and  p,  are  identical  to 
the  RBC  coefficients,  cp,  and  p^. 

Back-Scattering  by  Whole  Blood 

The  scattering  of  dense  particle  medium  is  a 
complex  theoretical  problem  which  has  already 
been  treated  by  Twersky  [8-10]  and  applied  by 
Shung,  Sigelmann  and  Reid  [4,5].    The  following 
simplified  approach  can  be  used.    Let  s'  =  qjuS 
be  the  functional  diametral  area  of  one  S.U., 


166 


where  s  is  the  diametral  area  of  one  S.U.  and 
qsu  is  a  dimensionless  number,  usually  greater 
than  1,  and  which  is  characteristic  of  the  S.U. 
packing.    For  a  spherical  S.U.  one  has: 
s'  =  qsu'fA^. 

We  are  interested  by  the  intensity  I^x  scatter- 
ed by  a  cylindrical  volume  element  having  the 
same  axis  as  the  ultrasonic  wave  with  height  Ax 
and  base  area  S  at  a  distance  x  from  the  probe 
(fig.  3). 


Probe 

If  1 


cylindrical  tube 
tilled  with  blood 


Fig.  3.    Simplified  diagram  of  the  sedimentation 
tube  with  the  echographic  probe. 

We  will  consider  in  this  volume  an  elementary 
slab  (S,  4A)  defined  by  the  two  sections  with 
distances  (x  +  2A)  and  (x  -  2A).    The  maximum 
number  of  S.U.  which  can  exist  in  this  slab  is 
2N  =  2S/s';  we  will  call  N  the  number  of  sites, 
each  composed  of  two  cells,  which  can  be  occupied 
by  0  or  1  or  2  S.U. 

Figure  4  shows  a  simplified  slab  with  7  sites 
occupied  in  such  a  way  that  the  ensemble  average 
of  the  hematocrit  is  4/7.    Eq.  (5)  only  applies 
to  the  sites  "f";  for  the  other  ones,  eq.  (5)  in- 
dicates that  the  scattered  intensity  is  zero. 


■2A 


X  +2A 


Fig.  4. 


Statistical  model.    The  slab  (x  -  2A, 
x  +  2A)  has  7  sites  with  an  occupation 
probability  of  4/7.    Only  the  "f"  emplace- 
ments have  a  favorable  configuration  for 
backscattering. 

By  a  simple  calculation  in  order  to  find  the 
probability  that  a  given  site  is  in  a  favorable 
configuration  (as  regards  back-scattering),  it  can 


easily  be  shown  that  the  number  of  sites,  Nf,  which 
have  a  favorable  configuration  in  an  elementary 
slab  2A  is: 


NH' (1  -  H'  ) 


(6) 


Assuming  that  Iq  and  Iq  do  not  depend  on  x  in 
the  volume  Ax  (this  point  is  justified  by  the  low 
values  of  the  attenuation  and  reflexion  coeffi- 
cients in  blood),  each  slab  2A  back-scatters  with 
the  intensity: 


I^^  =  NH'(1  -  H')Ip 


(7) 


As  there  are  Ax/2A  elementary  slabs  in  the  in- 
terrogated volume  (Ax,  S),  one  has: 


Iax 

Ax 


^(1 
2A 


H')I 


D 


This  expression  can  be  further  transformed  by 
replacing  N  by  S/s',  and  H'  by  PsuPhH.  using  eq. 
(5)  for  I[).    As  we  have  used  an  impenetrable 
spheres  model  with  a  cubic  packing,  the  factor 
Psu/Psu  is  equal  to  3/2.    Finally  one  obtains: 


Ax 


G(x)Kmv^p2H(l  -  P^Psu") 


(8) 


In  this  derivation  we  have  added  the  individual 
intensities  and  not  the  wave  amplitudes,  thus 
ignoring  the  interference  problem.    This  is  justi- 
fied in  a  first  approach  because  the  distribution 
of  the  scatterers  in  the  interrogated  volume  is 
random. 

Back-Scattering  Fluctuations 

Examination  of  eq.  (8)  shows  that  the  only 
fluctuating  quantities  are  H  or  H'  (the  packing 
factors  do  not  fluctuate  as  we  have  assumed  a 
spherical  model  for  the  aggregates.    It  is  also 
assumed  in  the  model  that  m  is  constant.    For  an 
elementary  variation  dH'  of  H'  in  a  slab  (2A),  the 
scattered  intensity  will  vary  according  to  eq. 
(7): 


^^2A  "  ^d"^'^  "  2H'  )dH' 


The  relative  mean  quadratic  fluctuation  F2A 
is  defined  as  follows: 


<(dl2A)'>^' 


2A 


<I2A> 


and  can  be  written: 


'^2A         <H'  (1  -  H'  )>  ^ 


2A 


a<(dH'  )2>i/2 


167 


If  one  transforms  dH'  into  dv,  one  obtains: 

As  in  the  initial  model  the  site  occupancy  proba- 
bilities are  independent,  the  S.U.  dis,tribution 
obeys  Boltzmann  statistics  and  <(dv)2>^  is  given 
by  [11]: 

<(dv)2>i/2  = 

(2AS)i'2 

Therefore  F2/\  can  be  written: 

=  a(2ASv^p^p^^m)V2<H.>i/2 

The  corresponding  fluctuation  coefficient  for 
the  blood  column  Ax  is  F^^: 

<(dl^^ 

Assuming  independent  fluctuations  for  the 
Ax/2A  elementary  slabs  in  Ax,  the  total  fluctua- 
tion is  given  by: 


to  =  2  IT  5.2  10^  s'l 

Po  =  1.078;  p  =  1.223 

Co  =  1550  m  s-i;  c  =  1610  m  s"! 

G(x)  =  1  (negligible  attenuation) 

m  =2  (slight  aggregation) 


—  theoretical  curve 
,  SHUNG  et  al  results 


10        M        Ml       40       50        60      H(  percent) 


'^.-C^Wlu^'y  (9) 


<H'>ife<(l  -  2H')2>ig 

<H'(1  -  H')>  Ax^ 

This  relation  shows  that  the  standard  deviation 
of  the  fluctuations,  given  by  F^^  *^ax^'  ''^  P'"'^" 
portional  to  m''/^.Ax''2. 

Discussion 

In  our  model,  the  interactions  between  RBC 
lead  to  monodisperse  spherical  aggregates.  In- 
deed, microscopic  observations  show  a  '^eversible 
aggregation  in  normal  and  pathological  blood  [12]. 
Our  picture  may  be  oversimplified,  as  these 
transient  aggregates  are  polydispersed  in  dimen- 
sion and  shape.    We  feel  however  that  introducing 
polydispersity  function  would  not  change  our 
treatment  fundamentally. 

Eguation  (8)  is  very  similar  to  a  previous  re- 
sult obtained  by  Shung  et  al.  [4]  using  Twersky's 
theory  [8-10].    However,  as  this  theory  predicts 
that  the  back-scattering  goes  through  a  maximum 
for  H  =  0.50,  Shung  et  al .  introduced  an  empiri- 
cal fitting  constant  as  their  experimental  re- 
sults showed  a  maximum  for  hematocrit  values  be- 
tween 0.25  and  0.30 

We  have  plotted  on  figure  5  their  experimental 
results  and  a  theoretical  curve  obtained  by  using 
eq.  (8)  and  the  following  values: 


Fig.  5.    Back  scattering  of  whole  blood  as  a  func- 
tion of  hematocrit:    correlation  between 
the  experimental  results  of  Shung  et  al . 
[4,5]  and  theoretical  values  calculated 
according  to  eq.  (8). 

Shung  et  al . ' s  scattering  coefficient  a  is  iden- 
tical to: 


I 

SAx 


The  quantitative  agreement  between  the  two  re- 
sults seems  satisfactory,  our  treatment  can  di- 
rectly explain  the  experimental  results  of  Shung 
et  al.  [4,5]. 

The  aggregation  dependence  of  the  scattered 
intensity  by  whole  blood  is  also  shown  in  eq.  (8) 
and  has  recently  been  demonstrated  by  Shung  and 
Reid  [13]. 

In  order  to  derive  the  fluctuation  coefficient, 
it  has  been  assumed  that  the  probability  of  find- 
ing a  S.U.  in  a  given  elementary  volume  is  propor- 
tional to  the  S.U.  number  concentration  and  in- 
dependent of  the  occupation  state  of  the  neighbour- 
ing elementary  volumes.    This  would  be  the  case 
for  pure  Brownian  motion.    However,  as  has  already 
been  pointed  out  by  Atkinson  and  Berry  [6],  the 
thermal  diffusion  of  RBC's  is  much  too  slow  (10^  s) 
to  travel  one  wavelength. 

Two  other  mechanisms  could  be  postulated.  The 
first  one  is  convection  streams  which  break  apart, 
associate,  rotate  and  translate  the  aggregates, 
so  that  the  occupancy  of  a  given  elementary  volume  | 
fluctuates  to  and  fro  in  the  occupation  probability. 
The  second  explanation  can  be  found  in  the  disso- 
ciation-association kinetic  of  the  S.U.    In  that  | 
case  it  is  m  which  would  fluctuate.    Further  ex-  I 


168 


periments  are  needed  in  order  to  assess  the  respec- 
tive importance  of  the  mechanisms. 

Summary 

The  ultrasound  back-scattering  of  blood  has  been 
shown  to  be  increased  when  the  erythrocyte  sedi- 
mentation rate  is  high  (fig.  1).    Moreover,  in  this 
case,  temporal  fluctuations  of  the  back-scattered 
intensity  have  also  been  demonstrated.    In  order 
to  explain  these  results,  a  simplified  theory  of 
ultrasound  back-scattering  by  blood  is  presented. 

The  following  model  of  the  blood  structure  is 
used:    the  erythrocytes  are  associated  to  form 
spherical  transient  aggregates  having  the  same  cell 
number,  m.    The  acoustic  wave  is  scattered  by  these 
clusters  called  "scattering  units". 

The  scattered  intensity  is  calculated  for  one 
kinetic  unit  taking  a  spherical  elastic  model, 
then  for  a  blood  unit  volume,  when  the  hematocrit 
H  is  very  low.    It  is  found  that  the  back-scattered 
intensity  is  proportional  to  H  and  to  m. 

When  H  is  not  negligible,  a  simplified  statis- 
tical theory  is  proposed.    The  main  result  is  that 
the  back-scattered  intensity  is  still  proportional 
to  m  but  that  it  passes  through  a  maximum  value 
when  H  gets  near  0.3. 

An  explicit  relation  between  the  mean  relative 
value  of  the  intensity  fluctuation,  H  and  m  is 
given. 

Therefore,  the  aggregation  state  of  blood  can 
in  principle  be  determined  through  ultrasound  scat- 
tering studies.    Further  theoretical  developments 
are  in  progress  so  as  to  take  into  account  the  dis- 
tribution of  the  number  of  erythrocytes  in  each 
kinetic  unit  and  the  variation  of  this  number  with 
H. 


[8]    Twersky,  V.,  On  scattering  of  waves  by  random 
distribution:    I.    Free  space  scatterer 
formalism,  J.  Math.  Phys.  3^,  700  (1962). 

[9]    Twersky,  V.,  On  scattering  of  waves  by  random 
distribution:    II.    Two  space  scatterer 
formalism,  J.  Math.  Phys.  3,  724  (1962). 

[10]    Twersky,  V.,  Acoustic  bulk  parameters  of 
random  volrame  distributions  of  small  scat- 
terers,  J.  Acoust.  Soc.  Am.  36,  1314  (1964). 

[11]    Landau-Lifchitz,  Statistical  Physics  (Mir 
ed.,  Moscow,  1967")^ 

[12]    Schmid-Schonbein,  M. ,  Gallasch,  G.,  Gosen, 
J.  v.,  Volger,  E. ,  and  Klose,  M.  J.,  Red 
cell  aggregation  in  blood  flow,  Klin.  Wschr. 
54,  149  (1976). 

[13]    Shung,  K.  K.  and  Reid,  J.  M.,  Ultrasonic 
detection  of  erythrocyte  aggregation,  29th 
ACEMB  Sheraton-Boston,  Boston,  Massachusetts, 
6-10  November  1976. 


References 

[1]    Hanss,  M. ,  Boynard,  M. ,  and  Perrin,  J. 

Erythrocyte  sedimentation  by  an  echographic 
method,  Biomedicine  25  (3),  81  (1976). 

[2]    Reid,  J.  M.,  Sigelmann,  R.  A.,  Nasser,  M.  G., 
and  Baker,  D.  W. ,  The  Scattering  of  Ultrasound 
by  Human  Blood.    Proceedings  of  the  8th  Inter- 
national Conference  in  Medicine  and  Biological 
Engineering  (1969). 

'j    [3]    Sigelmann,  R.  A.  and  Reid,  J.  M.  Analysis 
and  measurement  of  ultra-sound  backscatter- 
ing  from  an  ensemble  of  scatterers  excited 
by  sine  wave  bursts,  J.  Acoust.  Soc.  Am.  53, 
1351-1355  (1973). 

[4]    Shung,  K.  K. ,  Sigelmann,  R.  A.,  and  Reid, 
J.  M. ,  The  scattering  of  ultrasound  by  red 
blood  cells,  Applied  Radiology  77  (Jan., 
Feb.  1976). 

[5]    Shung,  K.  P.,  Sigelmann,  R.  A.,  and  Reid, 
J.  M.,  The  scattering  of  ultrasound  by 
blood,  I.E.E.E.  trans,  on  biomed.  Enain. 
BME  23  (6),  460  (1976).  ^ 

[6]    Atkinson,  P.  and  Berry,  M.  V.,  Random  noise 
in  ultrasonic  echoes  diffracted  by  blood, 
J.  Phys.  Math.  Nucl.  7  (11),  1293-1302  (1974). 

[7]    Morse,  P.  M.  and  Ingard,  K.  U.,  Theoretical 
Acoustics  (McGraw  Hill,  New  York,  1968). 


169 


CHAPTER  6 
TUMOR  DOPPLER  SIGNATURES 


171 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979) 


TUMOUR  DETECTION  BY  ULTRASONIC  DOPPLER  BLOOD-FLOW  SIGNALS 


P.  N.  T.  Wells,  M.  Halliwell,  R.  A.  Mountford, 
R.  Skidmore,  A.  J.  Webb  and  J.  P.  Woodcock 

Bristol  General  Hospital  and  Bristol  Royal  Infirmary 
Bristol,  United  Kingdom 


An  in  situ  cancer  is  avascular  and  harmless.    Vascul ari sation  is  necessary  before 
a  tumour  is  potentially  capable  of  rapid  growth.    Distinctive  continuous-wave  ultrasonic 
Doppler  signals  have  been  obtained  transcutaneously  from  malignant  breast  tumours. 
Similar  signals  have  been  detected  from  abdominal  malignancy,  by  means  of  a  pulsed 
Doppler.    These  signals  may  form  the  basis  of  a  practicable  method  of  screening  for 
breast  cancer,  and  of  improving  the  accuracy  of  cancer  diagnosis. 

Key  words:    Blood  flow;  breast  cancer;  diagnosis;  Doppler;  screening;  ultrasonics. 


1.    Tumour  Blood  Flow 

A  single  aberrant  cell  is  almost  certainly  the 
origin  of  every  solid  malignant  tumour.  This 
cell  divides  to  form  a  colony  of  malignant  cells, 
called  an  i_n  situ  cancer.    An  ijT_  situ  cancer  is 
avascular  and  harmless,  and  cannot  grow  beyond 
the  volume  limited  by  diffusion  of  nutrients  and 
wastes.    The  event  which  converts  an  in_  si tu 
cancer  into  a  "tumour"  is  apparently  the  release 
by  the  malignant  cells  of  a  diffusable  chemical 
substance,  called  the  "tumour  angiogenesis  fac- 
tor" [1]^.    Once  the  tumour  is  vascul ari sed  it  is 
potentially  capable  of  rapid  growth.    The  blood 
flow  in  normal  tissue  like  that  of  the  breast  is 
in  the  range  10  to  30  ml  min-i  kg-^  [2].  The  blood 
flow  in  lymphomas  is  about  380  ml  min~^  kg"^,  in 
differentiated  tumours,  140  ml  min"^  kg"^,  and  in 
anaplastic  tumours,  110  ml  min"^  kg"i  [3]. 

The  possibility  that  changes  in  blood  flow 
might  be  used  to  detect  malignant  tumours  is  dis- 
cussed in  this  paper,  and  some  preliminary  results 
are  presented.    Most  of  the  experimental  work  was 
done  with  breast  lesions,  because  of  their  acces- 
sibility and  the  potential  importance  of  the  method 
in  screening  for  breast  cancer. 

The  blood  vessels  which  serve  the  breast  are  il- 
lustrated in  figure  1.    The  mass  of  a  typical  fe- 
male breast  is  about  0.3  kg.    Assuming  that  the 
three  main  arteries  each  serve  one  third  of  the 
breast,  the  normal  flow  rate  in  each  artery  would 
be  around  1  to  4  ml  min"^.    A  malignant  tumour  with 
diameter  of  20  mm  would  require  a  flow  rate  of 
about  0.5  to  1.5  ml  min*^.    Therefore,  in  an  in- 
dividual, there  might  be  a  significant  difference 
between  the  flow  in  an  artery  supplying  a  tumour 
and  part  of  the  breast,  and  the  flow  in  the  cor- 
responding artery  serving  the  other  normal  breast. 
There  is  at  present  no  non-invasive  method  of  mea- 
suring blood  flow  volume  which  would  enable  this 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


Fig.  1.    Principal  blood  vessels  serving  the  right 
breast. 

to  be  tested.    It  may  be  that  a  change  might  occur 
in  the  pulsatility  index  of  the  flow  [4],  but  this 
would  be  difficult  to  establish  since  every  pair  of 
arteries  (serving  the  two  breasts)  would  need  to 
be  studied. 

In  biomedicine,  liquid  flow  through  small  ducts 
has  previously  been  recognised  in  backscattered 
ultrasonic  Doppler  frequency-shifted  signals  [5] 
in  two  situations.    The  first  of  these  is  in  the 
placenta  [6]  and  the  second,  in  the  fetal  lung  dur- 
ing breathing  [7].    Recently  it  has  been  reported 
that  characteristic  blood  flow  signals  are  asso- 
ciated with  neovascular  flow  in  malignant  tumours 
[8j  and  additional  data  are  presented  in  this  paper. 

2.    Breast  Investigations 

Breasts  were  examined  as  illustrated  in  figure 
2.    A  regular  continuous-wave  ultrasonic  Doppler 
blood-flow  detector  [5]  was  used.    The  hand-held 
probe  had  a  diameter  of  about  10  mm,  and  contained 
two  separate  8  MHz  transducers,  one  for  transmit- 
ting and  one  for  receiving.    The  transducers  were 


173 


HJNO-Hilll  HKOBl 


IRANSMIIIINC 
IfiiHSOUCtB 


Fig.  2.    Method  of  obtaining  ultrasonic  Doppler 
blood-flow  signals  from  the  breast. 


(a) 


(kHz) 


(b) 


fREOUEHC? 
Ikm) 


FREOUEdCt 
IkHi] 


aligned  so  that  the  volume  sensitive  to  motion  ex- 
tended axial ly  beyond  the  end  of  the  probe.  For 
each  patient,  the  Doppler  signals  obtained  from 
over  the  lesion  were  compared  with  those  from  the 
corresponding  site  on  the  opposite  breast.  Both 
breasts  of  each  normal  individual  were  thoroughly 
explored  to  search  for  Doppler  signals  which  were 
asymmetrical  and  characteristic  of  neither  normal 
arterial  nor  venous  flow. 

The  results  are  summarised  in  table  1.  Sono- 
grams of  the  signals  recorded  from  a  normal  mam- 
mary artery,  a  malignant  breast  tumour,  and  tis- 
sues at  the  corresponding  site  in  a  normal  breast, 
are  shown  in  figure  3. 

Table  1.    Ultrasonic  Doppler  signals  obtained  from 
breast  in  various  clinical  conditions. 


Confirmed  diagnosis^ 


Number  of  patients  with 
neovascularisation  signals 


None 

Weak 

Strong 

Normal 

2 

0 

0 

Mastitis 

1 

0 

0 

Mammary  duct 
pa  pi  1 lomatosis 

1 

0 

0 

Mammary  duct 
ectasia 

2 

0 

0 

Benign  mammary 
dysplasia 

2 

1 

0 

Cyst 

1 

2 

0 

Fi  broadenoma 

0 

1 

0 

Lymphoma 

0 

0 

1 

Carcinoma 

0 

0 

2 

Number  of  patients  =  16. 


3.     Investigations  of  Other  Tumours 

Signals  apparently  associated  with  malignant 
neovascularisation  have  been  detected  in  the  ab- 
domen.   Two-dimensional  scanning  was  used  to  locate 
a  pancreatic  tumour.    The  ultrasonic  beam  of  the 
scanner  probe  was  directed  through  the  tumour. 
The  probe  was  then  connected  to  a  2  MHz  pulsed 
Doppler  [5].    Blood  flow  signals  were  detected  when 
the  Doppler  was  range-gated  into  the  tumour. 


Fig.  3.    Sonograms  of  ultrasonic  Doppler  blood-flow 
signals  obtained  from  the  breast,  (a) 
Normal  individual:    branch  of  internal 
mammary  artery;  (b)  malignant  tumour;  and 
(c)  flow  at  site  corresponding  to  (b)  in 
other  breast  of  the  same  patient. 

4.  Discussion 
A.    Breast  Cancer 

Breast  cancer  is  the  most  common  malignancy  in 
Western  women.    In  the  United  Kingdom,  breast 
cancer  affects  at  least  80  in  100,000  females; 
it  kills  1  in  50  women,  1  in  75  women  between  the 
ages  of  35  and  60  years,  and  1  in  3,500  females 
of  all  ages  every  year.    Breast  cancer  is  the 
main  cause  of  all  deaths  among  women  between  the 
ages  of  40  and  44.    Extrapolating  from  1973  U.S. 
data  [9],  the  present  economic  cost  of  breast 
cancer  in  the  United  Kingdom  is  in  the  order  of 
$40  million  per  year  (and  $200  million  per  year 
in  the  United  States). 

Earlier  diagnosis  and  treatment  of  breast  cancer 
might  improve  the  survival  time  of  the  patient 
[10].    If  the  cancer  is  removed  whilst  it  is  still 
in  situ,  the  patient  is  cured.    The  next  step  in 
the  natural  progress  of  the  disease  is  when  the 
growing  cancer  begins  to  invade  the  surrounding 
tissue.    For  a  time,  the  cancer  cells  remain  local- 
ised.   If  diagnosed  and  treated  at  this  stage,  the 
5-year  survival  rate  is  around  85  percent.    If  un- 
treated, the  cancer  metastases,  with  regional  in- 
volvement of  the  lymph  nodes  which  drain  the  breast, 
and  the  5-year  survival  rate  falls  to  about  53  per- 
cent.   If  left  untreated,  metastases  occur  in  more 
distant  parts  of  the  body.    This  advanced  cancer  is 
virtually  incurable,  although  the  time  taken  for 
the  patient  to  die  is  variable. 

The  treatment  of  patients  with  early  breast 
cancer  is  more  successful  than  that  of  those  in 
whom  the  disease  is  advanced.    Breast  cancer  is 
usually  discovered  when  the  patient  herself  feels 
a  lump.    Even  in  a  "suitable"  breast,  the  smallest 
lump  which  can  be  detected  by  manual  palpation  is 
not  much  less  than  10  mm  in  diameter  [11].  The 
risk  of  the  presence  of  distant  metastases  from  a 
tumour  with  a  doubling  time  of  1.5  months  increases 
from  22  percent  when  its  diameter  is  1  mm,  to  43 
percent,  when  it  is  10  mm.    Once  a  lump  has  been 
discovered  in  a  breast,  present  diagnostic  proced- 
ures are  already  adequate,  although  there  may  be 


174 


disagreement  about  the  best  course  of  treatment  if 
the  lump  is  a  malignant  tumour.    The  possibility 
exists,  howt.^r,  that  diagnosis  of  pre-symptomatic 
lesions  by  an  effective  screening  procedure  might 
lead  to  a  reduction  in  breast  cancer  mortality  as 
a  result  of  earlier  treatment. 

Mammography,  either  conventional  or  xeroradio- 
graphic,  seems  to  be  the  best  contemporary  method 
of  detecting  early  breast  cancer  [12].  Unfortunate- 
ly, however,  mammography  as  a  screening  procedure 
confers  no  benefit  on  women  who  are  well  and  under 
the  age  of  50  [13].    In  these  women  only  19  percent 
of  cancers  would  not  have  been  detected  without 
mammography,  and  so  the  benefit  (which  has  to  be 
set  against  the  risk  of  radiation-induced  cancer 
as  a  consequence  of  mamography  [14])  is  small. 
Even  if  technical  developments  were  to  result  in  a 
reduction  in  x-ray  exposure  to  an  acceptable  level, 
the  logistics  of  interpreting  the  vast  number  of 
mammograms  which  would  be  obtained  in  a  screening 
programme  pose  apparently  insoluble  problems  of 
manpower,  boredom  and  expense. 

Thermography  has  been  used  to  study  breast 
tumours  [15]  but  initial  enthusiasm  for  the  method 
has  declined.    Thus,  in  a  series  [16]  of  2523 
volunteers,  1  patient  out  of  4  who  developed  cancer 
within  18  months  of  the  examination  was  detected  by 
thermography;  344  had  abnormal  or  suspicious  scans, 
but  no  abnormality.    These  and  other  results  are  so 
poor  that  the  method  (at  least  when  used  alone) 
seems  to  have  no  role  in  screening  for  breast 
cancer. 

Conventional  two-dimensional  pulse-echo  ultra- 
sonography is  capable  of  producing  images  of  the 
breast  of  quite  high  resolution  [17-20].  Un- 
fortunately, the  normal  breast  is  a  complex  ir- 
regular structure,  the  recognition  without  prior 
knowledge  of  small  tumours  is  very  unreliable,  and 
there  seems  to  be  no  way  of  automating  the  analysis 
of  the  vast  number  of  scans  which  would  result  from 
even  only  a  modest  screening  programme.    The  re- 
sults reported  in  the  present  paper  indicate  that 
characteristic  Doppler  signals  may  be  a  cancer- 
specific  ultrasonic  tissue  signature.    They  might 
form  the  basis  of  a  practicable  breast  cancer 
screening  method. 

B.    Other  Tumours 

Cancer-specific  Doppler  signals  could  have  an 
important  place  in  the  interpretation  of  the  in- 
adequate data  which  sometimes  results  from  con- 
ventional diagnostic  tests  such  as  radiography, 
echography,  scintigraphy,  and  computerised  tomo- 
graphy.   These  opportunities,  which  are  quite 
distinct  from  screening,  include,  for  example, 
the  differentiation  of  small  solid  and  cystic 
renal  lesions,  and  the  differentiation  of 
ophthalmic  tumours  and  organised  haematomas. 

References 

[1]    Folkman,  J.  and  Cotran,  R.  ,  Relation  of 
Vascular  Proliferation  to  Tumor  Growth,  in 
International  Review  of  Experimental  Pathology, 
C.  W.  Richter  and  N.  A.  Epstein,  eds..  Vol.  16, 
pp.  207-48  (Academic  Press,  New  York,  1976). 

[2]    Woodcock,  J.  P.,  Theory  and  Practice  of  Blood 
Flow  Measurement  (Butterworths ,  London,  1975). 


[3]  Mantyla,  M. ,  Kuikka,  J.,  and  Rekonen,  A., 
Regional  blood  flow  in  human  tumours  with 
special  reference  to  the  effect  of  radio- 
therapy, Br.  J.  Radiol.  49,  335-8  (1976). 

[4]    Woodcock,  J.  P.,  The  Significance  of  Changes 
in  the  Velocity/Time  Waveform  in  Occlusive 
Arterial  Disease  of  the  Leg  in  Ultrasonics  in 
Medicine,  L.  Pi  1  i pczyiiski ,  ed.,  pp.  243-50 
(Polish  Scientific  Publishers,  Warsaw,  1970). 

[5]    Wells,  P.  N.  T. ,  Biomedical  Ultrasonics 
(Academic  Press,  London,  1977). 

[6]    Hunt,  K.  M.  ,  Placental  localization  using  the 
Doptone  foetal  pulse  detector,  J.  Obstet. 
Gynaec.  Br.  Commonw.  76^,  144-7  (1969). 

[7]    Boyce,  E.  S.,  Dawes,  G.  S.,  Gough,  J.  D.,  and 
Poore,  E.  D.,  Doppler  ultrasound  method  for 
detecting  human  fetal  breathing  in  utero,  Br. 
Med.  J.  2,  17-8  (1976). 

[8]    Wells,  P.  N.  T.,  Halliwell,  M.,  Skidmore,  R., 
Webb,  A.  J.,  and  Woodcock,  J.  P.,  Tumour 
detection  by  ultrasonic  Doppler  blood-flow 
signals.  Ultrasonics  (to  be  published). 

[9]    Scitovsky,  A.  A.  and  McCall,  N.,  Economic 

Impact  of  Breast  Cancer  in  Frontiers  of  Radia- 
tion Therapy  and  Oncology,  J.  M.  Vaeth,  ed.. 
Vol.  11,  pp.  90-101  (Karger,  Basel,  1976). 

[10]    Brinkley,  D.  and  Haybittle,  J.  L.,  A  15-year 
follow-up  study  of  patients  treated  for 
carcinoma  of  the  breast,  Br.  J.  Radiol.  41 
215-21  (1968). 

[11]    Slack,  N.  H.,  Blumenson,  L.  E.,  and  Bross, 
I.  D.  J.,  Therapeutic  implications  from  a 
mathematical  model  characterizing  the  course 
of  breast  cancer.  Cancer,  N.Y.  24,  960-71 
(1969). 

[12]    Feig,  S.  A.,  Shaker,  G.  S.,  Schwartz,  G.  F. , 
Patchefsky,  A.,  Libshitz,  H.  I.,  Edeiken,  J., 
Nerlinger,  R. ,  Curley,  R.  F.,  and  Wallace, 
J.  D. ,  Thermography,  mammography  and  clinical 
examination  in  breast  cancer  screening. 
Radiology  122.  123-7  (1977). 

[13]    Strax,  P.,  Venet,  L.,  and  Shapiro,  S.,  Value 
.  of  mammography  in  reduction  of  mortality  from 
breast  cancer  in  mass  screening.  Am.  J. 
Roentg.  117,  686-9  (1973). 

[14]   ,  Radiation-Induced  Breast  Cancer, 

Br.  Med.  J.  I,  191-2  (1977). 

[15]    Williams,  K.  L.,  Lloyd  Williams,  F.  J.,  and 
Handley,  R.  S.  ,  Infra-red  thermometers  in 
the  diagnosis  of  breast  disease.  Lancet  2, 
1378-81  (1961). 

[16]    Hitchcock,  C.  R.,  Hickok,  D.  F. ,  Soucheray, 
J.,  Moulton,  T.  ,  and  Baker,  R.  C. ,  Thermo- 
graphy in  mass  screening  for  occult  breast 
disease,  J.  Am.  Med.  Ass.  204,  419-22  (1968). 


175 


[17]    Wells,  P.  N.  T.  and  Evans,  K.  T.  ,  An  immer- 
sion scanner  for  two-dimensional  ultrasonic 
examination  of  the  human  breast.  Ultrasonics 
6,  220-8  (1968). 

[18]    Jellins,  J.,  Kossoff,  G.,  Reeve,  T.  S.,  and 
Barraclough,  B.  H.,  Ultrasonic  gray  scale 
visualization  of  breast  disease,  Ul trasound 
Med.  Biol.  U  393-404  (1975). 

[19]    Baum,  G.,  Ultrasound  mammography.  Radiology 
122,  199-205  (1977). 

[20]    Kobayaski,  T. ,  Gray-scale  echography  for 

breast  disease.  Radiology  122,  207-14  (1977). 


176 


CHAPTER  7 
PROPAGATION  THROUGH  BONE  AND  SKULL 


177 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


A  THEORY  RELATING  SONIC  VELOCITY  TO  MINERAL  CONTENT  IN  BONE 


Sidney  Lees 

Forsyth  Dental  Center 
Boston,  Massachusetts    02115,  U.S.A. 

and 

Carel  L.  Davidson 

University  of  Amsterdam 
Amsterdam,  The  Netherlands 

Bony  tissues  consist  primarily  of  mineral  hydroxyapati te  (HAP)  crystallites  em- 
bedded in  a  matrix  of  a  much  softer  material,  collagen.  Currey  and  others  suggested 
that  bone  is  a  two-phase  composite,  like  mineral  filled  plastics,  but  the  known  laws 
of  mixtures  give  at  best  a  crude  approximation  of  the  observed  elastic  properties  of 
bone.  We  have  investigated  the  problem  by  measuring  the  ultrasonic  velocity  in  min- 
eral filled  particulate  composites  as  a  function  of  the  mineral  concentration. 

It  was  found  that  the  ultrasonic  velocity  can  be  predicted  for  some  mineral  filled 
plastics  by  applying  the  Reuss  formalism  to  the  longitudinal  elastic  modulus,  indicat- 
ing that  these  are  families  of  Reuss  solids  in  some  sense.    Other  mineral  filled 
plastics  do  not  seem  to  obey  this  rule,  the  ultrasonic  velocity  being  greater  than 
predicted.    Bone  appears  to  belong  to  this  latter  class  of  particulate  composites. 

The  literature  indicates  that  in  certain  situations  the  mineral  filler  affects 
the  plastic  matrix,  giving  the  plastic  constituent  a  higher  modulus.    A  maximum 
longitudinal  modulus  is  ultimately  attained  so  that  two  Reuss  formalism  bounds  can 
be  obtained,  the  lower  one  calculated  from  the  modulus  of  the  unfilled  plastic,  the 
upper  using  the  maximum  modulus.    The  ultrasonic  velocities  of  a  system  of  fluorapa- 
tite  (FAR)  filled  epoxy  was  found  to  lie  between  such  bounds. 

Currey  and  his  successors  assumed  bone  collagen  has  the  same  elastic  properties 
as  tendon  collagen  and  that  these  are  invariant  with  respect  to  contained  HAP.  The 
literature  shows  the  contrary,  that  bone  collagen,  even  when  demineral ized ,  is  more 
highly  cross  linked  than  any  other  collagen.    Moreover,  the  literature  indicates  that 
HAP  crystallites  are  chemically  bonded  to  the  collagen  molecules. 

It  is  postulated  that  bone  collagen  is  stiffened  because  HAP  crystallites  form  on 
the  intermolecular  cross  links,  encasing  them  and  making  their  effective  lengths  very 
short.    It  is  shown  that  the  sonic  velocity  for  bone  can  be  bounded  by  two  Reuss 
formalism  curves  in  the  same  manner  as  for  FAP  epoxy. 

Keywords:    Bone;  collagen;  crosslinking  modification;  curve! i nki ng ;  hydroxyapatite; 
sonic  velocity. 


1.    Bony  Tissues 

Bony  tissues  are  essentially  a  mineral, 
hydroxyapatite  (HAP),  embedded  in  an  organic  ma- 
trix, mostly  collagen.    HAP  is  an  hexagonal 
crystallite  which  in  bone  and  dentin  is  usually 
less  than  100  nm  in  any  dimension.    The  shape 
and  size  distribution  of  the  crystallite  in  bone 
has  been  studied  for  many  years  with  inconclusive 
results,  except  that  they  are  small.    Certain  x- 
ray  diffraction  studies  indicate  the  crystallites 
are  needlelike,  others  that  they  are  platelets. 
Electron  micrographs  show  platelets,  but  the 
needle  form  is  not  thereby  excluded.    There  is 
reason  to  believe  both  forms  are  present  de- 
pending on  the  site  of  formation.    There  is  also 


a  strong  contention  that  a  significant  fraction 
of  the  mineral  is  amorphous  [1]^. 

Collagen  is  a  generic  term  for  a  class  of 
protein  that  make  up  much  of  the  body  tissue. 
Bone  collagen  differs  from  other  collagens  in 
the  body  because  it  is  so  insoluble,  indicating 
a  high  degree  of  intermolecular  cross  linking. 
The  organic  part  of  bone  constitutes  about  65 
percent  by  volume  of  the  tissue  of  which  95  per- 
cent is  collagen.    The  mineral  occupies  35  per- 
cent of  the  volume.    Since  collagen  is  such  an 
important  body  constituent,  its  chemistry  has 
been  studied  intensively  for  many  years  and  is 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


179 


still  a  major  field  of  biochemistry.    Very  re- 
cently the  sonic  and  elastic  properties  of  some 
types  of  collagen  have  begun  to  be  studied  be- 
cause of  the  need  to  determine  the  presence  of 
collagen  in  tissues  and  to  identify  its  extent. 

All  collagens  are  highly  structured  materials 
in  a  multilevel  hierarchal  order.    The  molecular 
weight  of  the  basic  collagen  molecule,  defined 
as  tropocollagen,  is  approximately  300,000 
daltons,  which  is  a  very  large  molecule  even  in 
organic  chemistry.    The  molecule  is  much  like  a 
piece  of  spaghetti  having  a  wet  diameter  of 
1.5  mm  and  a  length  of  about  300  nm.    It  is  dif- 
ficult to  draw  to  scale  and  all  figures  in  this 
paper  are  purely  schematic.    As  noted,  bone  col- 
lagen differs  from  other  body  tissue  collagens 
by  an  extensive  network  of  intermolecular  cross 
links  that  render  it  insoluble  in  even  the  most 
potent  of  solvents.    On  the  other  hand,  tendon 
collagen  is  reported  to  be  mostly  lacking  in 
intermolecular  bonds  but  having  many  hydrogen 
bonded  intramolecular  links,  so  that  it  can  be 
dissolved  and  reconstituted  readily.    While  other 
tissues  like  arteries  and  joints  calcify,  there 
is  reason  to  think  that  the  special  structural 
characteristics  of  bone  collagen  cause  it  to 
calcify  in  a  unique  manner  [2]. 

2.    Two  Phase  Mi neral -Fi 1  led 
Plastic  Composites 

It  is  the  contention  of  this  paper  that  the 
mechanical  properties  of  bone  can  only  be  under- 
stood in  the  light  of  the  chemical  interrelation- 
ship between  the  mineral  crystallites  and  bone 
collagen.    Previously  Currey  [3],  Welch  [4]  and 
Katz  [5]  have  considered  bone  to  be  a  mechani- 
cally mixed  two  phase  mineral-filled  polymer 
composite.    Their  concept  yields  a  crude  rep- 
resentation of  the  variation  of  elastic  modulus 
with  mineral  content,  but  the  detailed  correla- 
tion is  not  good.    Currey  [6]  even  suggested 
that  the  crystallites  are  much  longer  than  ob- 
served in  order  to  attribute  a  fibrous  mineral 
structure  to  bone  to  account  for  the  observed 
elastic  and  strength  properties. 

We  believe  that  the  difficulties  of  our  pred- 
ecessors are  due  to  several  causes.    The  elastic 
modulus  of  collagen  is  not  well  known  and  there 
has  been  no  distinction  among  the  various  types. 
Previous  investigators  used  estimates  for  tendon 
fibers  rather  than  bone  collagen.    Except  for 
Mason's  [7]  sonic  velocity  measurement  on  kan- 
garoo tail  tendon  fibers,  the  estimates  of  the 
elastic  modulus  were  based  on  low  strain  rate 
tests.    Secondly,  there  was  no  attempt  to  examine 
the  chemical  structure  of  collagen  or  the  effect 
of  the  structure  on  the  properties  of  bone.  Col- 
lagen was  considered  a  structureless  homogeneous 
isotropic  continuum  and  the  short  range  order  of 
body  tissue  collagen  was  ignored.  Thirdly, 
there  is  little  understanding  even  at  this  date 
how  the  HAP  crystallites  are  laid  down  despite  a 
long  and  intensive  investigation  over  many  years 
by  many  investigators.    Fourth,  the  estimates  of 
the  elastic  modulus  of  bone  have  been  mostly 
those  obtained  by  standard  stress-strain  testers. 
Bone  is  vi scoelasti c .    Strain  rates  and  the  magni- 
tude of  the  strain  in  conventional  stress-strain 
testing  causes  hysteresis  and  often  permanent 
distortion.    In  this  paper  we  use  only  ultrasonic 
velocity  measurements  where  the  amplitude  of  dis- 


placement is  of  the  order  of  an  angstrom  and  the 
period  is  much  less  than  any  relaxation  time  of 
the  medium. 

3.    Ultrasonic  Measurements 

It  was  the  availability  of  measurements  of  the 
ultrasonic  velocity  of  several  hard  tissues  that 
led  to  the  elastic  theory  presented  here.  Con- 
ventional procedures  for  measuring  elastic  prop- 
erties are  difficult  because  bone  is  a  visco- 
elastic  material  and  the  data  must  include  the 
rate  of  change  of  strain  as  well  as  the  strain 
itself.    Moreover,  the  sample  is  significantly 
affected  by  the  test  process,  so  that  repeated 
tests  are  not  consistent.    Such  a  situation  re- 
quires further  interpretation  to  make  sense  of 
the  data  which  is  particularly  difficult  where 
the  material  (bone)  varies  widely  due  to  its 
biological  origin.    Ultrasonic  velocity  measure- 
ments, on  the  other  hand,  can  be  repeatedly  per- 
formed on  the  same  sample  with  an  uncertainty 
that  depends  mostly  on  the  measurement  technique 
and  not  on  the  sample.    The  contribution  of 
biological  variation  can  be  treated  separately 
apart  from  the  measurement  process. 

There  is  a  need  to  relate  the  elastic  prop- 
erties obtained  at  high  strain  rates  with  the 
more  conventional  techniques.    Papadakis  [8]  has 
designed  and  tested  an  infrasonic  resonator  for 
testing  plastics  at  about  one  hertz  but  which 
he  says  can  be  increased  to  1000  hertz.    It  will 
be  useful  to  find  the  sonic  velocity  measurements 
in  bone  at  these  low  frequencies,  but  the  limita- 
tions imposed  by  the  inhomogeneities  of  biologi- 
cal material  will  require  considerable  modifica- 
tion of  the  Papadakis  equipment. 

4.    Rule  of  Mixtures 

It  is  reasonable  to  consider  bone  to  be  a  two- 
phase  mineral-filled  composite  as  Currey  did, 
but  it  is  necessary  to  know  the  relationship  by 
which  the  elastic  properties  of  the  composite 
can  be  calculated  from  those  of  the  constituents. 
A  number  of  rules  of  mixtures  have  been  proposed 
for  estimating  the  elastic  moduli  of  the  compos- 
ite but  each  has  limited  applicability.  General- 
ly they  work  reasonably  well  when  the  properties 
of  the  two  constituents  are  nearly  alike  but 
when  there  is  a  tenfold  or  greater  ratio  in 
density,  the  mixing  rules  do  not  seem  to  apply. 

A  number  of  mineral  filled  plastics  were 
studied  to  determine  a  relationship  between 
longitudinal  sonic  velocity  and  mineral  con- 
tent.   The  first  series  included  tungsten- 
filled  vinyl,  crystabolite-filled  polymethyl- 
methacrylate, and  crystabolite-filled  epoxy. 
It  was  discovered  that  one  of  the  known  mix- 
ing rules  did  apply.    Reuss  postulated  an  ex- 
pression of  the  form: 


1  _   Vj_  V2. 

K  "  Ki  K2 


(1) 


where 


^1 ' 


volume  fraction  of  the  plastic 
consti  tuent 

volume  fraction  of  the  mineral 
constituent  =  1  -  Vj 
elastic  modulus  of  composite 
elastic  moduli  of  constituents. 


180 


In  this  application  the  longitudinal  modulus 
was  used  rather  than  the  bulk  and  shear  moduli 
which  was  Reuss's  intention.  Since 


K 


k  +  4/3 


(2) 


where  k  =  bulk  modulus  and  u  =  shear  modulus, 
another  value  for  the  longitudinal  modulus  can 
be  calculated  from  Reuss's  expression  applied 
first  to  the  bulk  modulus  and  then  to  the  shear 
modulus,  but  it  will  be  quite  different  from  the 
value  yielded  by  eq.  (1)  for  the  longitudinal 
modulus  directly  inserted.    Consequently,  the 
expression  in  eq.  (1)  is  designated  the  Reuss 
formal i  sm. 

The  sonic  velocity  is  calculated  as  usual 
from  the  relation: 

c  =  K/p  (3) 

where  p  =  Vjpj  +  P2V2  and  p  =  density. 

Figures  1,  2  and  3  taken  from  Lees  and  David- 
son [9]  show  how  well  Reuss's  formalism  applies 
to  these  examples.    It  is  not  obvious  why  and  we 
do  not  have  an  explanation.    It  is  characteristic 
of  these  figures  that  the  sonic  velocity  de- 
creased, or  at  best  does  not  increase  much,  as 
the  mineral  content  increases  on  the  left  hand 
side  of  the  curve,  even  though  calculations  show 
the  longitudinal  modulus  of  the  composite  is  in- 
creasing.   It  was  found  that  the  density  in  this 
regime  increases  much  more  rapidly  than  the 
elastic  modulus.    When  Vj  is  large  and  K2  is  ten 
times  Kj  the  first  term  on  the  right  side  of  eq. 
(1 )  dominates. 


1- 


VOIGT 


REUSS 


+  EXPERIMENTAL  VALUE 


0      10     20     30     40     50     60     70     80     90  100 
VOLUME  %  CRYSTABOLITE 

Fig.  2.    Polymethylmethacrylate-crystabol ite 
system  (from  Lees  and  Davidson  [9]). 


6 


0  H  1  1  1  1  1  1  1  1  1  — 

0      10     20     30     40      50     60     70     80     90  100 

VOLUME, PERCENT  TUNGSTEN 

Fig.  1.    Voigt-Ruess  bounds  for  vinyl -tungsten 
composite  system  with  experimentally 
determined  curve  (from  Lees  and 
Davidson  [9]). 


+  EXPERIMENTAL  VALUE 


1  - 


0  -I  1  1  1  1  1  1  1  1  1  

0      10     20     30     40     50     60     70     80     90  100 


VOLUME, PERCENT  CRYSTABOLITE 


Fig.  3.    Epoxy-crystabol i te  composite  system 
(from  Lees  and  Davidson  [9]). 


181 


5.    Modified  Conditions  for  Rule  of  Mixtures 

As  the  investigation  proceeded  a  new  kind  of 
mineral-filled  plastic,  f 1 uorapati te-f i 1 1 ed 
epoxy,  was  found  which  apparently  does  not  fol- 
low Reuss's  formalism.    Fluorapatite  (FAP) 
closely  resembles  HAP  but  unlike  HAP  it  can  be 
obtained  in  large  crystals.    Powdered  FAP  is 
much  more  crystalline  and  well  behaved  crystal lo- 
graphically  than  HAP.    Both  FAP  and  HAP  are 
surface  active  materials.    In  figure  4,  the 
lower  curve  shows  the  variation  of  sonic  veloci- 
ty of  the  composite  with  mineral  content  when 
the  longitudinal  modulus  of  the  unfilled  plastic 
is  taken  as  the  value  of  Kj .    It  may  be  observed 
that  all  the  measured  values  are  above  the  lower 
curve  and  that  there  is  some  scatter  among  them. 

Q  J  I  I  1  I  I  I  I  I  I  


7- 


LIMITING 
SONIC  VELOCITY 
BOUNDARY 


1  - 


0-1  1  1  1  1  1  1  1  1  1  

0      10     20     30     40     50     60     70     80     90  100 

VOLUME,  PERCENT  FLUORAPATITE 

Fig.  4.    Epoxy-f 1 uorapatite  composite  system 
(from  Lees  and  Davidson  [9]). 

The  only  difference  between  the  situations  of 
figures  3  and  4  is  the  replacement  of  one  mineral 
filler  by  another.    It  must  be  concluded  that 
sometimes  the  filler  can  influence  the  mechanical 
behavior  of  the  plastic  matrix.    Now  it  is  well 
known  that  polymers  become  stiffer  as  the  cross 
linking  density  increases  so  it  is  inferred  that 
FAP  must  cause  the  epoxy  to  increase  the  cross 
linking  density.    A  similar  situation  was  de- 
scribed by  Kumins  [10]  who  found  that  titanium 
dioxide  causes  an  increase  in  cross  linking 
density  in  epoxy,  apparently  by  a  factor  of  ten 


when  about  3  volume  percent  titanium  dioxide  was 
added.    Burhans  et  al .  [11]  showed  a  strong  de- 
pendence of  the  elastic  moduli  on  the  maximum 
cross  link  length,  so  that  either  an  increased 
cross  linking  density,  or  a  decreased  maximum 
cross  link  length,  or  both  can  increase  the 
elastic  modul i . 

In  order  to  see  how  this  could  be  applied  to 
the  FAP-epoxy  composite  system,  the  maximum 
value  for  the  compressive  modulus  reported  by 
Burhans  et  al .  was  used  as  a  basis  for  estimat- 
ing the  corresponding  maximum  value  for  the 
longitudinal  modulus.    A  new  Reuss  curve  was 
calculated  using  the  higher  modulus.    When  plot- 
ted on  figure  4  it  provided  an  upper  bound,  so 
that  all  the  experimentally  determined  velocities 
fell  between  two  Reuss  curves.    This  is  inter- 
preted to  mean  that  FAP  can  affect  the  cross 
linking  density  of  the  matrix.    The  scattered 
values  of  the  ultrasonic  velocity  indicates  a 
variability  in  the  cross  linking  density  from 
one  experiment  to  the  next.    The  samples  were  not 
always  prepared  the  same  way  and  the  changed 
conditions  could  possibly  modify  the  influence 
of  the  filler.    For  example,  the  samples  were 
compressed,  after  mixing,  in  an  attempt  to 
achieve  higher  mineral  content.    This  caused  a 
gray  exudate  to  ooze  from  the  mold,  the  gray 
color  being  attributed  to  the  very  fine  FAP 
particles.    But  these  fine  particles  are  most 
likely  to  have  maximum  influence  on  cross  link- 
ing density  because  of  their  increased  surface 
area . 

6.    Structure  of  Collagen 

It  is  now  possible  to  study  ultrasonic  wave 
propagation  in  bone  with  these  two  concepts  in 
mind.    First,  that  for  a  specific  two  phase 
mineral-filled  polymer  the  longitudinal  modulus 
of  the  composite  can  be  found  from  the  Reuss 
formalism  and,  second,  the  longitudinal  modulus 
of  the  matrix  can  be  affected  by  the  presence 
of  the  f i Her. 

While  collagen  is  a  polymer  and  bone  may  be 
considered  a  two-phase  mineral-filled  polymer, 
bone  is  not  formed  by  starting  with  a  mix  of 
mineral  and  monomer.    Collagen  is  laid  down 
first  and  then  it  becomes  mineralized  when  the 
HAP  crystallites  are  deposited.    When  bone  is 
demineral ized  the  collagen  matrix  can  be  re- 
covered as  a  rubbery  solid  having  the  form  and 
shape  of  the  original  bone.    However,  when  col- 
lagen is  removed  the  mineral  structure  left 
behind  is  a  weak  solid  that  easily  crumbles 
into  a  powder.    It  must  be  concluded  that  the 
collagen  forms  a  continuous  medium  but  the 
mineral  does  not.    The  mineral  fills  voids  in 
the  collagen. 

Collagen  has  been  studied  for  many  years  and 
much  is  now  understood  about  its  chemistry  and 
ul trastructure  but  many  unresolved  problems  re- 
main such  as  the  three-dimensional  order,  the 
chemical  character  of  the  i ntermol ecul ar  cross 
links  and  the  location  of  the  cross  links  (Rama- 
chandran  [13],  Gallop  and  Paz  [14],  Veis  [15]). 
Collagen  exhibits  many  levels  of  order  terminat- 
ing in  a  fibrous  structure.    The  successive 
levels  of  organization  can  be  understood  with 
the  aid  of  figure  5a  from  Lees  and  Davidson  [2]. 
The  smallest  element  is  the  a-helix  which 
spontaneously  combines  in  triplets  to  form  a 


182 


AXIS  OF 

EXTENDED 

CHAIN 

\ 


TRIPLE  AXESOF 
THREE  HELICES 


A      2.8  A 


3.1  A 


(a) 


(a)  The  a-helix  is  left  handed.  Each 
spot  is  a  residue  and  repeat  length 
is  3.1  A. 

(b)  The  a-helix  is  twisted  into  a  right 
handed  helix  which  reduces  repeat 
length  between  residues  to  2.8  K 
and  the  repeat  of  the  supercoil  is 
10  times  residue  length,  or  28  K. 

(c)  Three  a-helices  are  threaded  to  form 
a  single  tropocol lagen  (TC)  unit. 
(Adapted  from  Glimcher  and  Krane,  . 
1968  [12]). 

Fig.  5a.    The  tropocol lagen  unit  (from 
Lees  and  Davidson  [2] ) . 

superhelix  molecule,  the  tropocol 1 agen  (TC) 
unit.    In  turn  TC  forms  microfibril  ropes,  which 
in  their  turn  form  three-dimensional  fibrils. 
Fibrils  join  to  form  fibers  seen  in  tendon. 

1.    The  basic  element,  the  a-helix  chain,  has 
a  molecular  weight  of  about  100,000  and  several 
types  have  been  identified  depending  on  the 
amino  acid  composition.    The  aj  chain  has  1052 
amino  residues  of  which  1011  are  in  triplet  sets 
where  the  first  term  is  glycine.    The  N  terminus 
has  16  nonhelical  residues  beginning  with  an 


amino  (NH2)  group  while  the  C  end  has  25  non- 
helical  residues  terminating  in  a  COOH  group. 
Gallop  and  Paz  [14]  as  well  as  Hulmes  et  al .  [16] 
show  the  residue  map.    It  is  the  triplet  with 
first  term  glycine  that  defines  the  character- 
istic pattern  through  all  subsequent  hierarchal 
levels. 

2.  Three  a-helices  are  wound  into  a  super- 
helix  about  300  nm  long,  1.5  nm  diameter  (wet) 
and  a  molecular  weight  of  about  300,000  (Rama- 
chandran).    This  is  the  basic  molecule,  the  TC 
unit.    In  experiments,  the  a-helix  spontaneously 
links  up  to  form  collageneous  materials  which 
have  important  industrial  and  medical  uses  but 
the  product  is  not  the  ordered  collagen  of  bone 
and  tissue.    It  does  demonstrate  that  the 
hydrogen  bonded  state  between  the  a-helices  is 
stable  and  the  basis  for  interaction  between  TC 
units  in  the  higher  levels  of  structure. 

3.  TC  units  join  in  sets  that  exhibit 
"quarter  staggering"  in  tissue  as  illustrated 
in  figure  5b,  based  on  a  submolecular  length 
of  67  nm  or  234  residues,  the  D-length  (Hodge 
[17]).    A  TC  unit  is  4.4  D  units  long  and  packs 
into  a  structure  exhibiting  a  repeated  0.6  D  unit 
gap  between  TC  ends.    It  is  a  requirement  of 
Hodge's  scheme  that  no  two  quarter  stagger  gaps 
can  be  adjacent,  which  imposes  a  severe  restric- 
tion on  the  three-dimensional  structure  that  has 
not  yet  been  resolved  (Segrest  and  Cunningham 
[18]). 

4.  The  nonhelical  chains  at  each  end  of  the 
TC  unit  and  some  of  the  adjacent  glycine  coded 
triplets  are  probably  the  links  between  molecule 
ends  but  may  also  serve  to  link  helices  within 

a  single  molecule,  i.e.,  the  inter  and  intra- 
molecular links  are  similar.    However,  many  of 
the  intramolecular  links  must  be  hydrogen  bonds, 
while  much  of  the  intermol ecular  links  between 
parallel  molecules  must  be  chains  covalently 
bonded  as  side  chains  to  residues  in  the  back- 
bone of  the  molecule. 

5.  A  number  of  three-dimensional  structures 
have  been  proposed  for  the  microfibrils,  fibrils 
and  fibers.    We  use  Smith's  five  strand  rope  of 
TC  units  [19]  as  modified  by  Miller  and  Parry 
[20].    It  is  shown  in  figure  5c  as  Smith  pro- 
posed the  model  with  a  67  nm  axial  stagger,  72° 
aximuthal  displacement  between  axes  of  nearest 
neighbors  and  43  nm  hole  between  colinear  TC 
units.    It  incorporates  a  repeating  pattern 
after  5  x  67  nm  and  a  helix  with  a  five-fold 


1  — HwA 

2—  H  

3—  H  

4—  1  


0.6  D  HOLE  REGION 
N     I  C 


-A  V 


N  END  =  16  RESIDUE  LONG  NONHELICAL  END  +  TRANSITION  HELICAL  REGION 
C  END  =  25  RESIDUE  LONG  NONHELICAL  END  +  TRANSITION  HELICAL  REGION 
D  =  REPEAT  LENGTH  OF  234  RESIDUES 
(Adapted  from  Gallop  and  Paz,  1975  [14]) 

Fig.  5b.     "Quarter"  staggering  scheme  (from  Lees  and  Davidson  [2]). 


183 


40  A 


s 


CROSS  LINK 


Fig. 


5c.    Smith's  five  stranded  rope  model 
for  the  microfibril  (from  Lees 
and  Davidson  [2]). 


UNIT  CELL  80x  80A 


Fig.  5d.    Miller  and  Parry  model  of  the  structure 
and  packing  of  a  fibril  (from  Lees  and 
Davidson  [2]). 


screw  axis  (five  subunits  per  screw  turn  char- 
acterized by  the  holes).    The  pentagonal  unit 
is  4  nm  in  diameter  when  the  wet  TC  unit  is 
1.5  nm  diameter  and  the  lumen  is  1.1  nm. 

6.  Miller  and  Parry  indicated  that  the  five- 
strand  rope  must  be  twisted  to  produce  a  four- 
fold symmetrical  supercoil,  where  the  holes  are 
now  90°  apart,  but  there  is  still  a  pentagonal 
packing  of  the  strands.    The  colinear  TC  units  are 
inclined  about  2.5°  to  the  rope  axis  to  accom- 
modate the  twist. 

7.  Miller  and  Parry  have  deduced  a  fibril 
packing  structure  based  on  a  square  or  tetragonal 
cell  of  four  microfibrils  as  in  figure  5d.  While 
Miller  and  Parry  did  not  say  so,  the  pattern  of 
successive  helical  structures  suggests  that  the 
microfibrils  are  coiled  about  each  other  while 
maintaining  the  four  unit  cell. 

On  the  basis  of  this  description,  the  ultra- 
structure  of  collagen  can  be  schematically  rep- 
resented as  in  figure  6  where  the  microfibrils 


Si: 


INTERMOLECULAR 
CROSS  LINK 


Fig.  6.    Schematic  representation  of  the 

ultra-structure  of  collagen  showing 
intermolecular  cross  links. 


184 


are  tied  together  with  cross  links  of  organic 
chains.    The  TC  units  are  probably  very  stiff 
along  their  major  axis.    Enemoto  and  Krimm  [21] 
calculated  a  value  for  the  Young's  modulus  of  a 
very  similar  molecule,  polyglycine  II,  which 
they  found  to  be  41  GPa,  a  value  far  in  excess  of 
any  reported  value  for  collagen  or  even  for  bone 
(26  GPa).    Polyglycine  II  has  a  triple  helix 
hydrogen  bonded  molecule  and  closely  resembles 
collagen  in  its  structure  (Ramachandran  [13]). 
The  Young's  modulus  for  TC  is  probably  about  the 
same  value,  which  indicates  that  the  collagen 
structure  is  more  yielding  than  the  TC  units  be- 
cause the  cross  links  are  softer.  Moreover, 
the  long  skinny  TC  units  must  bend  quite  easily 
unless  they  are  severely  restrained,  which  makes 
collagen  tissue  softer  than  its  component  mole- 
cules. 

7.    Mineralization  of  Bone 

As  the  tissue  mineralizes,  HAP  crystallites 
begin  to  fill  the  voids  in  the  quarter  stagger 
gaps,  between  microfibrils  and  between  fibrils. 
The  loci  are  inferred  because  they  have  been 
sited  definitely  only  between  fibrils  (White  et 
al .  [1]).    It  is  our  contention  that  the  crystal- 
lites form  on  the  cross  links  until  they  grow  to 
a  size  that  fills  the  voids.    It  is  implied  that 
the  crystallites  start  with  the  cross  links  as 
the  nucleating  seeds,  which  further  implies  that 
the  stereochemistry  of  the  organic  chain  must  be 
favorable  to  the  hexagonal  habit  of  HAP.  There 
is  no  accepted  theory  to  explain  how  crystallites 


PLATELIKE 
CRYSTALLITE 


NEEDLELIKE 
^CRYSTALLITE 


Fig.  7.    Schematic  representation  of 

mineralized  bone  collagen  showing 
crystallites  embedding  the  cross 
1  inks. 


form  and  grow  in  tissue  but  some  workers  indicate 
that  the  process  is  mediated  by  a  sequence  of 
chemical  stages  rather  than  precipitating  from  a 
saturated  solution.    Whatever  the  process  it  en- 
tails a  mineralization  of  the  organic  cross  links 
until  the  crystallite  encases  the  links.  In 
figure  7  a  schematic  representation  of  the 
mineralized  microfibrils  shows  the  crystallites 
embedding  the  cross  links  and  effectively  short- 
ening them.    There  is  a  cross  link  from  one 
microfibril  to  a  crystallite  and  its  continua- 
tion on  the  other  side  connects  the  crystallite 
to  a  second  microfibril.    The  connections  are 
made  to  TC  units  in  the  strands  of  the  micro- 
fibrils, hence  to  side  chains  of  the  residues  on 
the  a-hel ix. 

The  effect  of  mineralization  in  bone  is  to 
shorten  the  cross  links  and  thereby  increase  the 
longitudinal  modulus  of  the  matrix.    There  may 
well  be  additional  cross  links  formed  at  the 
same  time  or  prior  to  mineralization,  but  the 
cross  linking  is  most  likely  mediated  by  a  dif- 
ferent chemical  process  than  that  for  mineraliza- 
tion.   The  stiffening  of  the  collagen  matrix  need 
not  necessarily  be  associated  with  a  greater 
cross  linking  density. 

8.    Ultrasonic  Velocity  of  Mineralized  Tissues 

In  order  to  apply  the  theory  it  has  been  neces- 
sary to  estimate  the  equivalent  isotropic  sonic 
velocity  for  hard  tissues.    Since  bone  and  dentin 
are  anisotropic  the  estimates  are  not  well  based 
and  the  ultimate  test  and  evaluation  of  this 
theory  requires  a  better  way  to  find  the  iso- 
tropic velocity  equivalent.    Alternatively,  it 
may  be  possible  to  extend  the  theory  to  deal  with 
the  anisotropy  as  Yoon  and  Katz  [22]  are  attempt- 
ing to  do  but  the  theory  is  still  incomplete. 

Figure  8  shows  representative  values  for  the 
longitudinal  sonic  velocity  of  bone,  dentin  and 
enamel.    There  are  four  bone  values,  one  for  nor- 
mal bovine  bone  by  Lang  [23],  and  three  by 
Abendschein  and  Hyatt  [24].    The  latter  include 
one  normal  human  bone  and  two  of  lower  density 
from  ill  people.    It  is  interesting  that  the  nor- 
mal human  and  normal  bovine  bone  values  coincide 
despite  the  quite  different  methods  for  making 
the  measurements. 

Unlike  the  mineral  filled  composites  we  do  not 
yet  have  good  values  for  the  elastic  properties 
of  the  constituents  of  bone,  particularly  bone 
collagen.    Several  Reuss  curves  were  calculated 
based  on  the  different  published  estimates  for 
the  elastic  modulus  of  collagen,  usually  the 
Young's  modulus  which  was  converted  to  the 
longitudinal  modulus  by  assuming  Poisson's  ratio 
to  be  0.35  (Katz  [5]). 

Three  Reuss  curves  are  shown  in  figure  8. 
The  lowest  bound  is  based  on  the  longitudinal 
modulus  suggested  by  Currey  [3]  and  Katz  [5], 
3.75  GPa.    The  intermediate  bound  is  calculated 
from  Mason's  value  [7]  measured  ul trasonical ly 
on  kangaroo  tail  tendon.    The  upper  bound  was 
obtained  by  using  Reuss 's  formalism  backward  on 
the  sonic  velocity  and  density  of  normal  bone. 

It  can  be  seen  from  figure  7  that  the  inter- 
mediate and  upper  curves  bound  the  experimental 
data,  much  like  the  two  bounding  curves  in  fig- 
ure 4.  The  lowest  bound  is  based  on  a  value  of 
longitudinal  modulus  based  on  low  strain  rate 
test  data,  a  technique  that  characteristically 


185 


3- 


1- 


INTERMEDIATE 
COLLAGEN 
K  MODULUS 


—[ — 
10 


—\ — 
20 


— I — 
30 


— I — 
40 


50 


— I — 
60 


70 


— 1 — 
80 


much  greater  than  isolated  collagen  because  in- 
timate association  with  the  mineral  raises  the 
collagen's  tensile  stiffness." 

References 

[1]      White,  S.  W.,  Hulmes,  D.  J.  S. ,  Miller,  A., 
and  Tinmins,  P.  A.,  Collagen-mineral  axial 
relationship  in  calcified  turkey  leg  tendon 
by  x-ray  and  neutron  diffraction.  Nature  266, 
421-425  (1977). 

[2]     Lees,  S.  and  Davidson,  C.  L.,  The  role  of 
collagen  in  the  elastic  properties  of  cal- 
cified tissues,  J.  Biomech.  10,  475-486 
(1977). 

[3]      Currey,  J.  D.,  Three  analogies  to  explain  the 
mechanical  properties  of  bone,  Biorheol ogy 
2,  1-10  (1964). 


-  [4] 


[5] 


90  100 


[6] 


VOLUME,  PERCENT  H YDROXYAPATI TE 


Fig.  8. 


Ultrasonic  velocity  in  mineralized 
tissues  (from  Lees  and  Davidson  [2]). 


yields  low  values  compared  to  ultrasonic  mea- 
surement techniques. 

The  data  represented  in  figure  8  can  only  be 
regarded  as  indicative  and  not  definitive.  In 
particular  the  value  of  enamel  cannot  be  taken 
as  properly  a  member  of  this  class  of  materials 
because  it  is  not  a  collageneous  tissue.  Enamel 
is  formed  by  a  totally  different  process  than 
that  for  bone  and  dentin.    The  HAP  content  of 
enamel  is  so  high  that  the  mineral  component  of 
eq.  (1)  dominates,  which  will  be  true  for  a  com- 
parable mineralized  collageneous  tissue.  How- 
ever, the  theory  developed  here  shows  that  it  is 
unlikely  to  have  so  much  mineralization  in  bony 
tissues. 

Acknowledgment 

S.  L.  is  pleased  to  acknowledge  support  from 
NSF  Grant  GH-42515  and  National  Institute  of 
Dental  Research,  NIH,  Research  Grant  Number 
ROl-DE-3992.    Some  of  the  experimental  data  was 
obtained  while  at  the  Department  of  Materials 
Science,  University  of  Amsterdam  with  funds  from 
the  National  Science  Foundation  and  the  Univer- 
sity of  Amsterdam. 


Note  added  in  proof: 

The  authors  have  recently  learned  that  some 
of  the  concepts  presented  here  were  surmised 
by  McCutchen  [25].    In  his  paper  he  says  "I  sug- 
gest that  collagen  is  the  prime  tension  carrier 
in  bone,  and  that  bone's  stiffness  modulus  is 


Welch,  P.  0.,  The  composite  structure  of 
bone  and  its  response  to  mechanical  stress, 
in  Recent  Advances  in  Engineering  Science, 
A.  C.  Eringer,  ed. ,  Vol.  5,  pp.  245-262 
(Gordon  and  Breach, 

Katz,  J.  L.,  Hard  tissue  as  a  composite 
material  -  I.  Bounds  on  the  elastic  be- 
havior, J.  Biomech.  4,  455-473  (1971). 

Currey,  J.  D.,  The  relationship  between 
the  stiffness  and  the  mineral  content  of 
bone,  J.  Biomech.  2,  477-480  (1969). 


[7]  Mason,  P.,  Viscoelasticity  and  structure 
of  keratin  and  collagen,  Kol loid ,  Z. ,  A. 
Polymere  202,  139-147  (1966T; 

[8]      Papadakis,  E.  P.,  Balanced  resonator  for 
infrasonic  measurement  of  Young's  modulus 
and  damping  in  flexure,  J .  Test  and  Eval . 
1,  126-132  (1973). 

[9]      Lees,  S.  and  Davidson,  C.  L.,  Ultrasonic 
measurement  of  some  mi neral -f i 1 1 ed 
plastics,  IEEE  Trans.  Sonics  and  Ultra- 
sonics SU-24,  222-225  (1977). 

[10]    Kumins,  C.  A.,  Long  range  effects  of 

polymer  pigment  interaction  in  the  solid 
state,  J.  Paint  Technology  Engineering  37, 
Pt.  1,  pp.  1314-1336  (1965). 

[11]    Burhans,  A.  S.,  Pitt,  C.  F.,  Sellers,  R.  F., 
and  Smith,  S.  G.  ,  High  performance  epoxy 
resin  systems  for  fiber-reinforced  com- 
posites.   Prelim.  Paper,  21st  Annual  Mtg. 
Reinforced  Plastics  Division,  Soc. 
Plastics  Ind.  (1965). 

[12]    Glimcher,  M.  J.  and  Krane,  S.  M. ,  in  Treati se 
on  Collagen,  B.  S.  Gould,  ed..  Vol.  2 
(Academic  Press,  New  York,  1968). 

[13]    Ramachandran ,  C.  N.,  in  Treatise  on  Col- 
1 agen ,  C.  N.  Ramachandran,  ed..  Vol.  1, 
Chap.  3  (Academic  Press,  New  York,  1967). 

[14]    Gallop,  P.  M.  and  Paz,  M.  A.,  Posttran- 

lational  protein  modifications  with  special 


186 


attention  to  collagen  and  elastin,  Physiol . 
Rev.  55,  418-487  (1975). 

[15]    Veis,  A.,  Collagen  Biosynthesis,  in  CRC 
Critical  Reviews  in  Biochemistry,  Vol.  2, 
p.  443  (The  Chemical  Rubber  Co.,  Cleveland, 
Ohio,  1974). 

[16]     Hulnies,  D.  J.  S.,  Miller,  A.,  Parry,  D.  A. 
D.,  Piez,  K.  A.,  and  Woodhead-Gal loway , 
J.  W. ,  Analysis  of  the  primary  structure 
of  collagen  for  the  origin  of  molecular 
packing,  J.  MoT .  Biol .  79,  137-148  (1973). 

[17]  Hodge,  A.  J.,  in  Treatise  on  Collagen,  C.  N. 
Ramachandran,  ed.,  Vol.  1,  Chap.  4  (Academic 
Press,  New  York,  1967). 

[18]    Segrest,  J.  P.  and  Cunningham,  L.  W. ,  Unit 
figril  models  derived  from  the  molecular 
topography  of  collagen,  Biopoly .  12,  825-834 
(1973). 

[19]    Smith,  J.  W.,  Molecular  pattern  in  native 
collagen.  Nature  219,  157-158  (1968). 

[20]    Miller,  A.  and  Parry,  D.  A.  D.,  Structure 
and  packing  of  microfibrils  in  collagen, 
J.  Mol .  Biol .  75,  441-447  (1973). 

[21]    Enemeto,  S.  and  Krimm,  S.,  Elastic  moduli  of 
helical  polypeptide  chain  structures, 
Biophys.  J.  2,  317-325  (1962). 

[22]    Yoon,  H.  S.  and  Katz,  J.  K. ,  Ultrasonic  wave 
propagation  in  human  cortical  bone,  J_.  Bio- 
mech ■  9_,  I.  Theoretical  considerations  for 
hexagonal  sysmmetry  407-412;  II.  Measurements 
of  elastic  properties  and  microhardness 
459-464  (1976). 

[23]     Lang,  S.  B.,  On  the  anisotropic  elastic  coef- 
ficients of  bone  and  results  on  fresh  and 
dried  bovine  bones.    IEEE  Trans.  Biomed. 
Engineering  17,  101-105  (1970). 

[24]    Abendschein,  W.  and  Hyatt,  G.  W.,  Ultra- 
sonics and  selected  physical  properties  of 
bone,  Clin.  Orthop.  69,  294-301  (1970). 

[25]    McCutchen,  C.  W. ,  Do  mineral  crystals 

stiffen  bone  by  straitjacketing  its  collagen? 
J.  Theor.  Biol .  51,  51-58  (1975). 


187 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ULTRASONIC  PROPERTIES  AND  MICROTEXTURE  OF  HUMAN  CORTICAL  BONE 


Hyo  Sub  Yoon  and  J.  Lawrence  Katz 

Center  for  Biomedical  Engineering 
Rensselaer  Polytechnic  Institute 
Troy,  New  York    12181,  U.S.A. 


The  wave  propagation  in  human  cortical  bone  (dried)  was  investigated,  using  an 
ultrasonic  pulse  transmission  method  at  room  temperature.    Firstly,  it  has  been  found 
that  the  symmetry  of  human  cortical  bone  is  consistent  with  the  hexagonal  system, 
based  on  the  ultrasonic  velocity  measurement  and  microscopic  observations.    The  five 
independent  elastic  stiffnesses  were  determined  at  5  MHz,  and  they  are  (in  GPa): 
Cii  =  23.4  ±  0.31,  C33  =  32.5  ±  0.44,  c^i,  =  8.71  ±  0.13,  Cjj  =  9.06  ±  0.38, 
Ci3  =  9.11  ±  0.55. 

Secondly,  this  study  shows  that  bone  filters  and  polarizes  ultrasonic  waves. 
Thirdly,  since,  in  a  piezoelectric  medium  such  as  bone,  the  wave  propagation  (or 
elastic  stiffnesses)  is  modified  by  the  piezoelectric  coupling,  the  piezoelectric 
contribution  or  "stiffening"  was  calculated  for  bone,  employing  the  piezoelectric 
and  dielectric  constants  of  bone  reported  in  the  literature.    Compared  with  the 
corresponding  values  of  two  well-known  piezoelectric  materials,  a-quartz  (single 
crystal)  and  "poled"  barium  titanate  ceramic  (polycrystal 1 ine ) ,  it  has  been  found 
that  the  piezoelectric  "stiffening"  in  bone  is  negligible.    Finally,  the  sound 
velocities  were  measured  over  the  frequency  range  of  1  to  5  MHz  for  the  transverse 
mode  and  of  2  to  10  MHz  for  the  longitudinal  mode.    For  all  the  eight  independent 
modes  the  ultrasonic  velocities  are  found  to  increase  with  increasing  frequency, 
implying  that  bone  is  viscoelastic  even  at  these  high  frequencies. 

Key  words:    Anisotropy;  dispersion;  elasticity;  human  bone;  microhardness ;  micro- 
structure;  piezoelectricity;  thermodynamics;  ultrasound;  visco- 
elasticity;  wave  propagation. 


1.  Introduction 

The  material  symmetry  is  an  important  concept 
for  interpreting  and  analyzing  the  structure- 
property  relationships  of  a  material,  from  both 
theoretical  and  practical  points  of  view.  This 
is  especially  true  when  one  has  to  deal  with  many 
tensorial  properties  of  a  multiphase-composite 
material  such  as  bones  and  teeth.    The  maximum 
use  of  symmetry  elements  in  the  medium  simplifies 
the  analysis  enormously  and  provides  a  clearer 
physical  picture  for  any  phenomenon  than  other- 
wise. 

It  is  known  that  bone  has  a  remarkably  well 
organized  structure,  consisting  mainly  of  the 
protein  (collagen,  not  well  crystallized),  the 
inorganic  phases  (crystalline  hydroxyapatite  and 
possibly  a  type  of  amorphous  calcium  phosphate) 
and  a  fluid  phase  (in  vivo).    Therefore,  one  can 
expect  that  several  crystal-physical  phenomena 
occur  in  bone.    It  has,  in  fact,  been  reported 
that  bone  is  both  pyroelectric  (first-rank 
tensor)  and  piezoelectric  (third-rank  tensor). 
When  employing  ultrasonic  waves  of  low  strain 
levels  much  less  than  10"^,  as  in  diagnostic 
ultrasound,  one  must  take  into  account  these 
interactions  between  thermal  and  electrical 
properties,  and  between  elastic  and  electrical 


properties,  respectively.    In  other  words,  it  is 
necessary  to  regard  bone  as  a  thermodynamic  sys- 
tem.   However,  the  situation  becomes  more  com- 
plicated for  a  more  realistic  case  than  the 
above  idealized  one  because  bone  appears  to  be 
viscoelastic  even  at  ultrasonic  frequencies. 

It  is  now  clear  from  the  above  information 
that  one  cannot  treat  each  physical  or  thermo- 
dynamic property  independent  of  the  others,  as 
has  been  done  in  the  past,  e.g.  ,  [l,2]i.  Instead, 
one  should  start  with  a  simple  situation  when 
studying  bone,  then  move  to  a  more  complex  model 
by  adding  one  or  more  independent  variables,  as 
was  followed  in  the  present  investigation  on 
human  compact  bone.    For  this  purpose  it  is  con- 
venient to  assign  hierarchal  levels  of  struc- 
tural organization  to  bone,  that  is,  molecular, 
ul trastructural ,  microscopic  and  macroscopic. 
The  wavelengths  of  ultrasound  employed  (3.7  x 
10" 1  to  2.2  mm)  are  of  the  order  of  magnitude 
appropriate  to  the  microscopic  level.    This  is 
the  level  at  which  conventional  metal lographic 
techniques  are  best  suited  to  study  the  micro- 
structure  or  microtexture  of  bone.    Note  that 
the  microtexture  of  human  bone  is  different  from 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


189 


that  of  bovine,  canine,  or  any  other  mammalian 
bone. 

The  simplest  situation  for  ultrasonic  studies 
of  bone  is  one  in  which  temperature  (room)  and 
frequency  (5  MHz)  are  kept  constant.    This  cir- 
cumstance corresponds  very  closely  to  the  situa- 
tion of  plane  wave  propagation  in  an  elastic 
medium  and  yields  a  set  of  elastic  stiffnesses 
for  bone.    As  the  next  step,  the  piezoelectric 
contribution  was  considered  for  the  ultrasonic 
wave  propagation  in  bone.    So  far,  bone  has  been 
treated  as  an  elastic  dielectric.    In  order  to 
determine  the  existence  of  a  frequency  dependence 
or  dispersion  of  the  ultrasonic  velocities  at 
room  temperature,  the  sound  velocities  were 
measured  over  the  frequency  range  of  1  to  5  MHz 
for  the  transverse  mode  and  of  2  to  10  MHz  for 
the  longitudinal  mode.    These  results  are  im- 
portant when  it  is  necessary  to  accurately  de- 
lineate ultrasonic  paths,  as  in  ultrasonic  ence- 
phalography, monitoring  of  fracture  healing,  and 
other  medical  and  dental  diagnoses  since  the  at- 
tenuation of  ultrasonic  waves  in  calcified  tissue 
is  substantially  larger  than  that  in  soft  tissue. 

2.    Theoretical  Background: 
The  Thermodynamics  of  Bone 

In  order  to  derive  the  relationships  between 
the  mechanical,  electrical  and  thermal  proper- 
ties of  bone,  i.e.  to  establish  the  thermo- 
dynamic aspects  of  bone,  let  us,  following  Voigt 
[3],  introduce  a  thermodynamic  potential  per 
unit  volume,  E,  as  a  function  of  ten  independ- 
ent variables:    stress  T^,  electric  field 
strength  E^-  and  entropy  a,  combined  with  the  ap- 
propriate coefficients,  elastic  stiffness  c^v^ 
dielectric  permittivity  e-jj,  specific  heat  C, 
piezoelectric  stress  constant    e-jy,  thermal 
stress  coefficient    q^  pyroelectric  constant  p-j 
and  absolute  temperature  0.  Thus: 


1        E,a    r    c    X    1        S,0    r.r.     ,     I  ^ 

^  c  '    S  S  +  ly  £  .  ■    Ei  E-i  +  y  -^-F 


(1) 


6  qt 


+  e^    EiS„  + 


IP    '  y  CE 


S,,a  + 


e  p-: 
 J\ 

cs 


EiO, 


where  summation  over  repeated  indices  is  implied, 
and  i  ,j  =  1,  2,  3;  M,  V  =  1,  2,  .  .  . ,  6.  A 
superscript  indicates  the  quantity  to  be  kept 
constant.    Here  magnetic  effects  are  not  includ- 
ed in  eq.  (1)  because  they  are  usually  very  small 
compared  to  electric  effects.    Also,  bone  is 
treated  as  an  anisotropic  linear  elastic  solid, 
i.e.  as  a  linear  elastic  dielectric,  rather  than 
as  either  metallic,  ferromagnetic,  or  ferro- 
electric solids.    That  is,  for  simplicity  the 
viscoelastic  properties  are  considered  separate- 
ly. 

Since  the  propagation  of  ultrasonic  waves 
through  a  medium  is  an  adiabatic  process  (isen- 
tropic),  eq.  (1)  is  simplified  to 


ic^'°  S  S 
2    pv     V  \ 


1  y 


EiSy 


(2) 


Differentiation  of  eq.  (2)  with  respect  to  Sp 
and  E-j,  respectively,  gives  the  well-known  equa- 


tions of  state  (or  constitutive  equations)  for  a 
piezoelectric  medium: 


T  =  c^'^  S 
^  yv 


eV 


i  /  T,a 


S,a 


(3) 
(4) 


which  now  define  stress  T^  and  electric  displace- 
ment D..    Note  that  the  coefficients  on  the  lead- 
ing diagonal  represent  the  principal  effects 
while  the  off-diagonal  coefficients  (equal  to 
each  other)  measure  the  coupled  effects. 

For  the  pulsed  ultrasonic  measurements  of 
thickness  vibration,  the  elastic  stiffness  ap- 
propriate for  the  vibration  would  properly  be 
designated  as 

Dn,Et,a 
^y  V  ' 

meaning  that  the  normal  component  of  the  electric 
displacement  and  the  transverse  components  of  the 
electric  field  are  constant  (usually  zero).  By 
combining  Newton's  second  law  of  motion  with  the 
constitutive  eqs.  (3)  and  (4)  and  substituting 
plane  wave  solutions  uj  (=  Uj  exp  i(a)t  -  k-x)) 
for  the  medium  of  density  p,  it  may  be  shown  that: 


3t2 


c . 


Dn,Et,a 

ijk?. 


ax  .3x, 


where : 


°n'^t'°  ^    Dn'EfO  ^    E,a  ^ 


'^i'jk?  "-yv 


with  m  being  fixed. 


e  p 
my.  "ly 


,S,a 
mm 


(5) 


(6) 


Note  that  the  directions  i,  n,  m  are  parallel  to 
one  another,  and  that 


Dn'^t'° 


D,o 


(7) 


is,  in  general,  different  from  c^^^  which  is 
given  by: 

D,a       E,o  ^iy^jv 

C  =    C  +   r   ' 

yv  yv  b,CT 

=  ij 

For  more  details  the  reader  is  referred  to 
references  [4-6]. 

3.    Experimental  Procedure 


A.    Specimen  preparation 

Bone  samples  were  cut  with  a  band  saw  out  of 
the  right  femoral  mid-diaphysi s  from  the  cadaver 
of  a  57-year-old  male  who  died  of  gastric  adeno- 
carcinoma.   Two  wel 1 -oriented  specimens  in  the 
shape  of  a  rectangular  parallelepiped  (0.5461  cm 
x  0.8933  cm  x  0.6233  cm  for  X1X2X3  cut  and 
0.4750  cm  x  0.5641  cm  x  0.7645  cm  for  45°  cut), 
were  then  cut  using  an  Isomet  low  speed  saw  with 
a  diamond  blade  10.16  cm  diameter  and  0.03048  cm 
thick.    Distilled  water  was  used  as  the  lubricant 


190 


and  coolant  during  cutting.    The  specimens  were 
oriented  using  a  straight  surface  in  the  medial 
quadrant,  which  is  always  parallel  to  the  bone 
axis  (X3  axis)  in  any  femur,  taking  the  axis 
along  the  radial  direction  and  the       axis  per- 
pendicular to  both,  thus  forming  a  right-handed 
rectangular  coordinate  system.    The  lateral 
quadrant  of  the  mid-di aphysi s  was  selected  for 
obtaining  the  specimens  because,  for  this  partic- 
ular bone,  this  quadrant  was  macroscopically 
more  uniform  than  was  the  rest  of  the  cross 
section. 

Conventional  metal lographic  techniques  were 
then  employed  for  grinding  and  polishing  the 
specimens.    In  order  to  prepare  a  pair  of  paral- 
lel surfaces  for  ultrasonic  measurements,  the 
specimen,  mounted  on  a  parallel -face  device  with 
double-stick  tape,  was  ground  under  a  flow  of 
cooling  water,  with  an  AB  Handimet  Grinder  on 
which  a  600  grit  silicon  carbide  paper  strip  was 
pasted.    This  was  followed  by  polishing  with  0.3 
pm  alumina  slurry  on  a  20.32  cm  diameter  bronze 
wheel  covered  with  wool  broadcloth.    Each  time 
after  grinding  and  polishing,  the  specimen  was 
cleaned  in  an  ultrasonic  cleaner  for  3  minutes. 
The  orientation  of  a  specimen  was  checked  by 
microscopic  observation  of  osteon  arrangement  in 
the  transverse  cross  section. 

Each  specimen,  wrapped  with  soft  tissue  papers, 
was  dried  very  slowly  to  avoid  undesirable  results 
such  as  cracks  and  distortions.    This  was  done  by 
keeping  the  specimen  in  a  glass  vial  whose  plastic 


lid  was  tightly  closed,  for  a  day,  followed  by 
placing  it  first  in  a  desiccator  for  10  days,  and 
subsequently  in  a  vacuum  oven  at  24  °C  and  533  Pa 
for  4  days.    Finally,  the  specimen  was  dried  in  a 
vacuum  line  equipped  with  both  a  mechanical  pump 
and  a  diffusion  pump  at  room  temperature  for  3 
days.    Before  proceeding  with  the  ultrasonic 
velocity  measurements,  the  moisture  content  of 
the  specimen  was  stabilized  by  keeping  it  in  air 
until  changes  in  the  specimen's  dimensions  were 
recorded  as  less  than  0.1  percent. 

B.    Micrographs  and  microhardness 
measurements 

From  the  same  mid-diaphysi s  of  human  bone, 
another  cylindrical  specimen  0.5  cm  long  was  pre- 
pared in  the  identical  manner  as  described  before. 
This  specimen  was  used  both  for  obtaining  micro- 
graphs (75X)  of  the  transverse  cross  section  and 
for  Vickers  (HV)  microhardness  measurements,  em- 
ploying a  Bausch  &  Lomb  Metal loqraph  and  a 
Kentron  Tester,  respectively.    A  typical  micro- 
structure  of  the  transverse  cross  section  is 
shown  in  figure  la.    Compare  it  with  that  of 
young  bovine  bone  shown  in  figure  lb.    The  hard- 
ness of  the  bone  specimen  was  measured  with  a 
200  g  load  at  locations  equally  spaced  radially 
on  the  four  quadrants:    anterior,  posterior, 
lateral  and  medial.    Care  was  taken  to  make  all 
indentations  away  from  Haversian  canals  or 
lacunae. 


r 


(a)  HUMAN 


(b)  BOVINE 


Fig.  1.    Microstructures  of  human  and  bovine  demora,  75X  (transverse  cross  section), 
(a)  human  and  (b)  bovine. 


C.    Ultrasonic  setup  and  measurements 

A  block  diagram  for  the  pulse  through-trans- 
mission technique  is  shown  in  figure  2.    For  each 
frequency  step,  a  pair  of  PZT-5A  piezoelectric 
transducers  (chromium-gold  plated  on  both  faces) 
was  used  to  transmit  and  receive  the  ultrasonic 
pulses.    For  the  specimen  holder,  a  pair  of 
parallel  faces  of  aluminum  buffer  disc  was  pre- 


pared by  hand-grinding  it  on  400-  and  600-grit 
silicon  carbide  paper  strips,  as  explained  pre- 
viously, and  by  polishing  with  a  3  pm  diamond 
compound;  the  disc  was  cleaned  ul trasonical ly 
after  each  step.    The  transducer  was  cemented  on 
one  face  of  the  buffer  disc  with  salol  (melting 
point  43  °C),  the  other  face  being  in  contact 
with  a  face  of  the  specimen  during  the  transit- 
time  measurements.    No  coupling  medium  was  used 


191 


RF  PULSED  OSCILLATOR 
(ARENBERC  P&- 65UCI 


PRECISION  ATTENUATOR 
lARENBERG  ATT-6931 


J2^_TRANSM  TRANSDUCER 
( PZT  5A I 


OSCILLOSCOPE 
ITEKTRONII  5431 


IZ,^   RECEIV  TRANSDUCER 
(PZT  5A) 


WIDE  BAND  AMPLIFIER 
(H  P  460  A) 


Fig.  2.    Block  diagram  of  pulse  transmission  method. 

between  the  buffer  disc  and  the  specimen  surface 
since  dried  bone  is  porous  and  the  specimen  volume 
was  very  small.    Instead,  the  acoustic  coupling 
was  formed  by  lightly  pressing  down  the  upper 
part  of  the  specimen  holder  with  a  screw.  For 
more  details  see  reference  [7]. 

Pulse  transit  (or  delay)  times  in  the  specimen 
were  measured  from  the  shift  of  the  pulse  posi- 
tions as  observed  on  the  horizontal  axis  of  the 
oscilloscope  with  and  without  the  specimen  be- 
tween the  aluminum  alloy  buffer  discs.  The 
horizontal  (time)  axis  of  the  oscilloscope  was 
calibrated  with  a  Time-Mark  Generator  180A.  The 
accuracy  of  the  pulse  transmission  method  was 
checked  by  measuring  the  transit  times  both  of 
AT  cut  quartz  oscillator  plates  of  known  frequen- 
cies, and  a  beryl  single  crystal  whose  values 


were  determined  previously  by  Yoon  and  Newnham 
[8]  using  a  pulse  superposition  method  [9]. 

4.    Experimental  Results  and  Discussion 

A.    The  homogeneity  and  microtextural 
symmetry  of  human  cortical  bone 

Figure  3  shows  that  the  bone  is  "intrinsically" 
homogeneous  within  the  limit  of  experimental  er- 
rors along  the  radial  direction  on  the  four 
quadrants.    Weaver  [10]  has  also  observed  a  wide 
zone  of  mid-cortical  bone  of  uniform  hardness  for 
autopsy  and  surgical  bone  specimens.  Kallieris' 
results  [11]  show  no  significant  differences  in 
hardness  between  fresh  human  bones  of  various 
structural  designs  in  different  layers  of  compact 
bones.    These  findings  justify  somewhat  the  use  of 
small  specimens  for  measurements  of  the  physical 
or  mechanical  properties  of  bone. 

Rauber  [12]  and  Dempster  and  Liddicoat  [13] 
have  shown  that  bone  has  different  physical  and 
mechanical  properties  along  and  perpendicular  to 
its  long  axis.    Therefore,  the  symmetry  (or 
pseudosymmetry )  of  the  microstructural  texture  of 
bone  may  be  either  orthorhombic,  tetragonal,  trig- 
onal, or  hexagonal.    Since  bone  is  an  opaque, 
microstructural  composite,  neither  optical  nor 
single  crystalline  x-ray  diffraction  techniques 
can  be  employed  to  determine  its  full  textural 
symmetry.    An  ultrasonic  wave  propagation  method 
can  provide  information  about  both  the  symmetry 
and  the  elastic  (and  viscoelastic)  properties  of 
either  crystalline  or  non-crystalline  materials. 
The  microstructure  on  the  transverse  cross  sec- 
tion of  bone  shows  a  pseudo-hexagonal  close- 


BOr 


40- 


ANTERIOR 


Relative  scale 


80 


r  40 


•     •      •  •     •     •  • 


POSTERIOR 


_L 


Relative  scale 


ci.  BOi- 


■r  40- 


LATERAL 


Relative  scale 


80 


40 


•       •  • 


MEDIAL 


Relative  scale 


Fig.  3.    Microhardness  (Vickers  Hardness)  of  human  femur  (dried). 
P  =  periosteum  and  E  =  endosteum. 


192 


packing  of  Haversian  systems  or  osteons.  Pre- 
viously, Katz  [14]  and  Katz  and  Ukraincik  [15] 
suggested  that  the  arrangements  of  osteons  and 
interstitial  lamellae  could  be  considered  to  be 
pseudo-hexagonal.    Thus,  the  tetragonal  and  trig- 
onal systems  can  be  eliminated.    The  ultrasonic 
measurements  have  shown  that  Cn  and  C22  or  c,^l^ 
and  C55  are  equal  to  each  other,  thereby  leaving 
only  hexagonal  symmetry  as  appropriate  for  bone. 

B.    The  elastic  stiffness  of  bone 

Table  1  shows  the  ultrasonic  velocities  to- 
gether with  their  standard  deviations  at  room 
temperature  and  5  MHz  along  various  directions 
in  the  human  femur  specimens.    The  eight  in- 
dependent velocity  measurements  provide  four  in- 
ternal cross-checks.    For  comparison  the  cor- 
responding sound  velocities  in  a  Durango  fluo- 
rapatite  single  crystal  [16]  are  included.  From 
these  velocities  and  the  independently  measured 
mass  density  p,  the  elastic  stiffnesses  of  bone 
were  calculated,  as  shown  in  table  2  where  the 
pseudo-single  crystal  elastic  stiffnesses  of 
hydroxyapatite  (HAP)  [15],  enamel  and  dentin 
[17]  are  included,  in  addition  to  the  elastic 
stiffnesses  of  fluorapatite  (FAP)  [16].    Note  in 
the  case  of  the  mineralized  tissues  that  the 
stiffnesses,  in  general,  increase  with  the  amount 
of  apatite. 


Table  1. 


Sound  velocities  in  human  compact  bone  and 
fluorapatite  crystal  at  room  temperature. 


Mode    Propagation    Displacement       Sound  velocity  (km/s) 

direction,      direction,    Fl uorapati te^    Human  femur 
N  U  (dried) 


aL 
aT 
yL 

TT3 

45L 

45Th 

45T„ 


[001] 
[001] 
[100] 
[100] 
[100] 

[A  oi] 


[001] 
[100]  or  [010] 
[100] 
[010] 
[001] 


1/2  "i] 

[0  -1  0] 


7.586 
3.635 
6.842 
3.978 
3.638 

7.020 
3.811 
4.033 


4.18 
2.16 
3.55 
1 .98 
2.17 

3.86 


0.03 
0.01 
0.02 
0.01 
0.02 

0.03 


2.06  ±  0.01 


2.24  ±  0.02 


See  reference  [16]. 


^  1 


1        1  1 

aL 

1              1  1 

1  1 

■--.^,^^^451 

_ 

  yL 

-aT 

51  —  

45Tv 

 0  

YT3- 

 <i 

1111 

45Th 

1        1  1 

 >; 

YTh 

1  1 

0      10     20     30     40      50     60  70 
Colatitude,  (f)  (degree) 


80  90 


Fig.  4.    Angular  dependence  of  the  elastic  stiff- 
nesses of  human  bone. 

while  the  degenerate  transverse  mode  aT  along  the 
bone  axis  (X3)  is  separated  out  into  the  two 
branches,  45Tv  ^  YT3  and  ISJ^  ^  yT^^. 

In  a  hexagonal  medium  there  exists,  in  general, 
three  pure  mode  directions  (a,  g,  y)  along  each  of 
which  one  purely  longitudinal  and  two  purely 
transverse  modes  of  propagation  occur,  forming  a 
mutually  orthogonal  set.    Of  these  the  a  and  y 
directions  (see  table  1)  are  known  from  the  sym- 
metry of  the  hexagonal  medium.    The  6  direction, 
which  lies  on  a  conical  surface  between  the  X3 
axis  and  its  perpendicular,  is  given  in  terms  of 
the  elastic  stiffnesses  of  the  material.    It  has 
been  found  that  no  6  direction  exists  in  bone. 
It  may  be  that  in  bone  the  structural  arrangement 
of  the  collagen  fibrils  and  fibers  and  the  osteons 
may  not  allow  any  direction  of  high  symmetry  for 
the  6  pure  mode.    This  can  explain  the  observed 
phenomena  that  bone  polarizes  and  filters  the 
ultrasonic  waves. 


Table  2.    Elastic  stiffnesses  of  calcified  tissues  and  apatite 
single  crystals  at  room  temperature. 


Cuu    Fluorapatite  Hydroxyapatite Enamel  Dentin  Bone 
(GPa)        [16]  [15]  [17]     [17]  (dried) 


150 

5 

137 

115 

37 

0 

23 

4  ±  0.31 

C33 

185 

0 

172 

125 

39 

0 

32 

5  ±  0.44 

42 

51 

39 

6 

22 

8 

5 

70 

8 

71  +  0.13 

C12 

48 

8 

42 

5 

42 

4 

16 

6 

9 

06  ±  0.38 

62 

2 

54 

9 

30 

0 

8 

7 

9 

11  ±  0.55 

p(g/cm5) 

3 

2147 

3 

17 

2 

9 

2 

2 

1 

86 

In  figure  4  are  plotted  the  elastic  stiff- 
nesses of  bone  as  a  function  of  angle  4)  from  the 
X3  axis.    This  shows  how  the  elastic  stiffnesses 
are  interrelated  to  the  orientation  in  a  plane 
parallel  to  the  X3  axis.    In  other  words,  the 
longitudinal  modes  aL,  45L  and  yl  are  related, 


C.    The  piezoelectric  "stiffening" 

Following  Fukada  and  Yasuda  [18]  and  Fukada 
[19],  point  group  6^  was  chosen  as  the  piezo- 
electric class  of  symmetry  for  bone,  which  is 
also  consistent  with  the  pyroelectric  measure- 
ments on  bovine  bone  by  Lang  [1].    In  table  3  is 
summarized  the  piezoelectric  contribution  to  the 
elastic  stiffnesses  of  bone,  except  for  the  two 


^While  it  is  true  that  bone  does  not  exhibit 
crystalline  symmetry  in  the  usual  sense,  as  would 
be  demonstrated  by  x-ray  diffraction  analysis,  6 
is  the  appropriate  point  symmetry  to  describe  the 
microstructural  level  of  the  organization  of  bone 
in  that  it  is  consistent  both  with  the  morpho- 
logical observations  and  with  the  ultrasonic  data 
within  the  precision  of  the  measurements. 


193 


Table  3.    Piezoelectric  corrections  to  elastic 
stiffnesses  of  bone. 


Mode 


-Dn,Et,a 


^Dn.Et.a  _  E,a 


JJV_ 


aL 

^33 

aT 

yL 

TTh 

YT3 

cE 

45Th 

2 

633 
£33 


2 

ei5 


-11 


£11  +  £33 


1.9  X  10-8 

0 

0 

0 

3.9  X  10-7 
2.3  X  10-6 


mixed  modes  (45L  and  45Ty),  employing  the  piezo- 
electric constants  for  horse  femur  [19]  and  the 
permittivities  of  bovine  bone  by  Gundjian  and 
Chen  [20]  with  the  correction  for  frequency  de- 
pendence following  Liboff  and  Shamos  [21].  Note 
that  the  piezoelectric  "stiffening"  in  bone  is 
very  small  and  may  therefore  be  neglected  within 
the  experimental  errors.    Table  4  compares  the 
values 


(cD'°  -  cE'<')/cE.« 


of  the  typical  piezoelectric  materials  a-quartz 
and  polarized  barium  titanate  ceramic,  with  the 
corresponding  values  of  bone  (see  2.  Theoretical 
background). 

Table  4.    Comparison  of  the  piezoelectric  contribution 

to  the  elastic  stiffnesses  of  bone,  a-quartz  and 
and  BaTi03  ceramic. 


E,a\  /  E.o 


Bone  (room  temp.)  a-Quartz  (20  °C)  BaTiO,  ceramic  (25  °C) 
Point  group,  6      Point  group,  32     Point  group,  6  mm 
(x  10-')  (x  io-2)  (x  10-2) 


11 

0.18 

0.86 

0 

93 

33 

0.19 

0 

17 

0 

44 

82.0 

0.072 

28 

0 

12 

0.46 

-11.0 

2 

0 

13 

0.56 

0 

-  8 

4 

14 

-  0.99 

66 

0 

1.9 

0 

D.    The  frequency  dependence  of 
ultrasonic  velocities 

Figure  5  shows  how  the  eight  sound  velocities 
in  dried  human  bone  change  with  frequency  at  room 
temperature.    In  all  cases  the  ultrasonic  veloci- 
ties are  found  to  increase  with  increasing  fre- 
quency.   All  the  transverse  modes  exhibit  almost 


^  4.0- 


3.5 


3.0 


2.4 


45L 


J  L 


1_L 


1     23456789  10 
Frequency,  MHz 


2.2- 


2.0- 


1.8 


45T, 


J  


Fig.  5. 


1          2         3         4  5 
Frequency,  MHz 

Frequency  dependence  of  sound  velocities 
in  human  bone  (dried). 


similar  dispersion  behavior,  while  one  of  the 
longitudinal  modes  (yL)  has  a  steeper  slope  than 
the  rest  of  modes . 

Brillouin  [22]  summarizes  the  explanations  of 
the  geometric  dispersion  due  to  a  periodic  or 
discrete  lattice,  which  were  originated  by  Cauchy 
and  refined  by  Powell  and  Kelvin.    This  book  also 
includes  Brillouin's  own  research  on  the  subject. 
More  recently,  Sutherland  and  Lingle  [23]  report- 
ed the  geometric  dispersion  of  acoustic  waves  by 
an  elastic-elastic  composite  of  tungsten  wires 
embedded  in  an  aluminum  matrix  over  the  incident 
frequency  range  of  0.63  to  8.57  MHz.    They  also 


194 


reported  the  existence  of  pass-  and  forbidden- 
bands,  in  addition  to  the  shift  of  the  incident 
frequency  upon  propagating  through  the  material. 
In  their  case  the  phase  velocity  always  decreases 
with  increasing  frequency  within  each  pass  band. 

For  a  linear  viscoelastic  solid,  e.g. ,  poly- 
methyl  methacrylate,  Asay,  Lamberson  and  Guenther 
[24]  reported  on  the  viscoelastic  dispersion  of 
ultrasonic  waves,  in  which  the  phase  velocity 
increases  with  increasing  frequency  and  approaches 
a  plateau.    These  results  can  be  explained  by  re- 
laxation phenomenon  of  "molecular"  units  (see 
Gross  [25]). 

As  mentioned  earlier,  bone  is  a  hierarchal 
composite  on  many  levels  of  structure.    Of  prin- 
cipal concern  here  is  the  microstructural  level 
of  organization.    In  this  case  the  osteons  be- 
have as  stiff  hollow  elastic  fibers  ensheathed 
by  a  compliant  viscoelastic  material,  mainly  col- 
lagen and  protein  polysaccharides. 

It  is  clear  that  much  additional  data  are  need- 
ed with  respect  to  ultrasonic  attenuation  in  bone 
over  a  wide  frequency  range  as  well  as  spectral 
analysis  of  both  transmitted  and  reflected  pulses. 
In  addition  it  would  be  valuable  to  perform 
studies  on  synthetic  composites  of  collagen  and 
hydroxyapatite  in  various  combinations  in  order 
to  obtain  dispersion  data  from  controlled  com- 
posites.   Still  from  the  modelling  described 
above,  it  is  reasonable  at  present  to  explain  the 
dispersion  of  the  phase  velocities  in  human  com- 
pact bone  in  terms  of  its  viscoelastic  components. 

Acknowledgment 

Contribution  No.  94  from  the  Laboratory 
for  Crystallographic  Biophysics;  this  work 
was  supported  by  US  PHS  through  NIDR  Grant 
No.  5Tl-DE-n7-14. 

References 

[1]    Lang,  S.  B.,  Pyroelectric  effect  in  bone 
and  tendon.  Nature  212,  704-705  (1966). 

[2]    Lang,  S.  B.,  Ultrasonic  method  for  measur- 
ing elastic  coefficients  of  bone  and  re- 
sults on  fresh  and  dried  bovine  bones, 
IEEE  Trans.  Bio-Med.  Engng.  17,  101-105 

TTW:    ^  — 

[3]    Voigt,  W.,  Lehrbuch  der  Kristal Iphysik 

(B.  G.  Teubner,  Leipzig,  1910);  reprinted 
with  an  additional  appendix,  1928  (Johnson 
Reprint  Corp.,  New  York,  1966). 

[4]    Cady,  W.  G. ,  Piezoelectricity  (Dover  Pub- 
lications, New  York,  1964). 

[5]    Yoon,  H.  S.  and  Katz,  J.  L.,  Ultrasonic 
wave  propagation  in  human  cortical  bone  - 
I.    Theoretical  considerations  for 
hexagonal  symmetry,  J.  Biomech.  9^,  407- 
412  (1976). 

[6]    Yoon,  H.  S.  and  Katz,  J.  L. ,  Ultrasonic 
wave  propagation  in  human  cortical  bone  - 
III.    Piezoelectric  contribution,  J_. 
Biomech.  9,  537-540  (1976). 


[7]    Yoon,  H.  S.  and  Katz,  J.  L.,  Ultrasonic 
wave  propagation  in  human  cortical  bone  - 
II.    Measurements  of  elastic  properties 
and  microhardness ,  J.  Biomech.  9,  459- 
464  (1976). 

[8]    Yoon,  H.  S.  and  Newnham,  R.  E.,  The  elastic 
properties  of  beryl ,  Acta  Cryst.  A29,  507- 
509  (1973). 

[9]    McSkimin,  H.  J.,  Pulse  superposition 

method  for  measuring  ultrasonic  wave  velo- 
cities in  solids,  J.  Acoust.  Soc.  Am.  33, 
12-16  (1961). 

[10]    Weaver,  J.  K.,  The  microscopic  hardness  of 
bone,  J.  Bone  Joint  Surg.  48A,  273-288 
(1966). 

[11]    Kallieris,  D.,  Hartemessungen  an  frischen 
menschlichen  Knochen,  Z.  Rechtsmed.  68, 
164-170  (1971). 

[12]    Rauber,  A.  A.,  Elasticitat  und  Festigkeit 
der  Knochen  (Engelmann,  Leipzig,  1876). 

[13]    Dempster,  W.  T.  and  Liddicoat,  R.  T., 

Compact  bone  as  a  non-isotropic  material. 
Am.  J.  Anat.  9^,  331-362  (1952). 

[14]    Katz,  J.  L. ,  Anisotropic  elastic  properties 
of  calcified  tissues,  lADR  Program  and 
Abstracts  of  Papers,  p.  71,  Abstract  No. 
73  (1971)  (49th  General  Session,  Intern. 
Assoc.  Dent.  Res.,  Chicago,  Illinois). 

[15]    Katz,  J.  L.  and  Ukraincik,  K. ,  On  the 

anisotropic  elastic  properties  of  hydro- 
xyapatite, J.  Biomech.  4,  221-227  (1971). 

[16]    Yoon,  H.  S.  and  Newnham,  R.  E.,  Elastic 
properties  of  f 1 uorapati te ,  Am.  Mineral. 
54,  1193-1197  (1969);  Yoon,  H.  S.,  Elastic 
properties  of  fluorapatite  and  beryl,  Ph.D. 
Thesis  (Pennsylvania  State  University, 
University  Park,  PA,  1971). 

[17]    Lees,  S.  and  Rollins,  F.  R.  ,  Jr.,  Anisotropy 
in  hard  dental  tissues,  J.  Biomech.  5^,  557- 
566  (1972). 

[18]    Fukada,  E.  and  Yasuda,  I.,  Piezoelectric  ef- 
fects in  collagen,  Jap.  J.  Appl.  Phys.  2» 
117-121  (1964). 

[19]    Fukada,  E.,  Mechanical  deformation  and 

electrical  polarization  in  biological  sub- 
stances, Biorheology  5_,  199-208  (1968). 

[20]    Gundjian,  A.  and  Chen,  L.  L.,  Standardiza- 
tion and  Interpretation  of  the  Electro- 
mechanical Properties  of  Bone,  IEEE  Trans. 
Biomed.  Engng.  21,,  177-182  (1974T; 

[21]    Liboff,  A.  R.  and  Shamos,  M.  H. ,  Solid  State 
Physics  of  Bone,  in  Biological  Mineralization, 
I.  Zipkin,  ed..  Chap.  14,  pp.  335-395  (Wiley, 
New  York,  1973). 

[22]    Brillouin,  L.    Wave  Propagation  in  Periodic 
Structures  (Dover  Publications,  New  York, 
1953).  ^ 


195 


[23]    Sutherland,  H.  J.  and  Lingle,  R.  ,  Geometric 
dispersion  of  acoustic  waves  by  a  fibrous 
composite,  J.  Composite  Mater.  6_,  490-502 
(1972). 

[24]    Asay,  J.  R.,  Lamberson,  D.  L.,  and  Guenther, 
A.  H.,  Pressure  and  temperature  dependence 
of  the  acoustic  velocities  in  polymethyl- 
methacrylate, J.  Appl.  Phys.  40,  1768-1783 
(1969). 

[25]    Gross,  B.,  Mathematical  Structure  of  the 
Theories  of  Viscoelasticity  (Hermann  & 
Cie. ,  Paris,  1953). 


196 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525  (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


ATTENUATION  AND  DISPERSION  OF  ULTRASOUND  IN  CANCELLOUS  BONE 


James  E.  Barger 

Bolt  Beranek  and  Newman  Inc. 
Cambridge,  Massachusetts  02138 


Measurements  of  the  insertion  loss  and  insertion  phase  shift  for  ultrasound 
transmitted  through  sections  of  cancellous  bone  from  human  skull  are  reported  as 
functions  of  frequency  over  the  range  extending  from  0.3  MHz  to  3.0  MHz.  The 
frequency  dependence  of  insertion  loss  and  of  phase  speed  are  both  found  to  be 
caused  principally  by  scattering  of  sound  by  the  blood  and  fat  filled  interstices 
in  the  bone  matrix.    Independent  scattering  measurements  made  at  all  observation 
angles  confirm  the  scattering  phenomenon.    It  is  concluded  that  both  the  high 
attenuation  and  the  significant  dispersion  at  frequencies  above  about  1  MHz  will 
limit  the  ability  to  characterize  brain  tissue  by  its  backscatter  at  these  high 
frequencies.    Also,  most  reported  sound  speeds  in  cancellous  bone  have  been  calcu- 
lated from  the  time-of-f 1 ight  of  broad-band  pulses,  and  are  therefore  group  speeds. 
These  group  speeds  will  exceed  the  phase  speeds  by  about  15  percent  in  magnitude. 


Keywords:    Attenuation;  dispersion;  skull  bone;  sound  speed;  ultrasound. 


1.  Introduction 

The  objective  of  this  research  was  to  measure, 
as  a  continuous  function  of  frequency  in  the  range 
extending  from  250  kHz  to  2.5  MHz,  both  the  inser- 
tion loss  and  the  phase  speed  of  ultrasound  in 
cancellous  bone.    These  measurements  are  especially 
Important  as  design  data  for  ultrasound  systems 
that  are  to  characterize  tissue  through  skull  bone, 
which  comprises  two  outer  tables  of  ivory  bone  and 
an  inner  layer  of  cancellous  bone,  called  diploe. 
This  diploe  layer  is  known  to  contribute  most  of 
the  ultrasonic  attenuation,  as  well  as  pulse  wave- 
form distortions,  at  diagnostic  system  frequencies 

Successful  tissue  characterization  involves  fre- 
quency spectrum  analysis,  over  a  rather  broad  band 
of  frequency,  of  the  sound  scattered  by  a  range- 
gated  segment  of  tissue  [2].    The  attenuation  that 
occurs  when  scattered  sound  is  transmitted  through 
diploe  can  change  the  spectrum  in  the  band  in  which 
signal  amplitude  exceeds  noise  amplitude  by  an 
amount  sufficient  to  permit  spectrum  identification 
Also,  the  pulse  waveform  distortions  owing  to  fre- 
quency-dependent sound  speed,  also  called  disper- 
sion, can  substantially  reduce  the  range  resolution 
of  the  broadband  scattered  sound. 

The  experiments  reported  in  this  paper  were  all 
performed  on  cancellous  bone  samples  obtained  by 
removing  the  ivory  tables  from  a  9.5-mm  thick  sec- 
tion of  human  skull  that  had  been  fixed  for  about 
six  months  in  buffered  formalin. 

The  experimental  apparatus  is  described  schemati 
cally  in  figure  1.    A  pulse  generator  applies  a 

^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


PULSE 
GENERATOR 


PDPll  /40 


RF  OUTPUT 


TRIGGER 


INPUT 


Focused 
Transducer 


Diploe 
Section 


Glass-Walled 
tonk 


Ultrosonic  Probe  at  Focus 


Fig.  1.    Schematic  diagram  of  experimental  ap- 
paratus for  measuring  insertion  loss  and 
phase  through  bone  samples. 

broadband,  high-voltage  pulse  to  a  focused  trans- 
ducer.   This  transducer  generates  at  its  focus  a 
pulse  having  a  peak  pressure  of  about  eight  bars 
and  a  bandwidth  (defined  at  10  dB  down  points)  of 
about  300  kHz  to  2.5  MHz.    The  pulse  is  captured 
and  digitized  at  20  MHz  and  subsequently  transmit 
ted  to  a  PDPll/40  computer  for  spectral  analysis. 

The  ultrasonic  waveforms  are  measured  at  the 
focus  of  the  transducer  with  a  small  piezoelectri 


197 


disc  probe  3  mm  in  diameter  and  0.3  rrm  thick.  The 
disc  is  positioned  with  its  major  response  axis 
colli  near  with  the  axis  of  the  ultrasonic  beam. 
The  disc  diameter  is  equal  to  the  diameter  of  the 
sound  beam  at  about  its  6-dB  down  points,  so  that 
the  probe  output  voltage  is  proportional  to  the 
sound  pressure  averaged  over  the  sound  beam. 

The  insertion  loss  and  phase  of  the  diploe 
sample  are  functions  of  the  Fourier  coefficients 
T(f)  of  the  pulse  at  the  focus  beyond  the  diploe 
sample  and  of  the  Fourier  coefficients  S(f)  of  the 
pulse  also  measured  at  the  focus,  but  without  the 
diploe  sample.    The  insertion  loss  in  dB  and  the 
insertion  phase  angle  are  defined  in  eq.  (1).  Both 
functions  are  calculated  at  19.5-kHz  intervals  of 
frequency  by  the  computer. 

IL  =  -20  log  |T/S|  (1) 

(})EargT-argS. 

The  phase  speed  of  sound,  Cj,  in  the  diploe 
sample  can  be  calculated  from  the  frequency  ratio 
of  the  insertion  phase  angle,  ((i/f  =  (j)'.    The  inser- 
tion phase  angle  is  the  increase  in  phase  of  sound 
transmitted  through  the  diploe  layer  less  the  in- 
crease in  phase  of  sound  transmitted  through  a  water 
layer  having  the  same  thickness,  h,  as  the  diploe 
layer.    This  increase  is  expressed  in  eq.  (2), 
where  41  is  the  phase  angle  in  degrees  and  Cq  is  the 
phase  speed  of  sound  in  water. 

<}.  =  360  f  h  (CT'  -  c-i)     .  (2) 

Solving  for  the  phaee  speed  in  diploe,  we  have 

ci  =  Cq{1  +  <l)'  c^/SeOh)-'  .  (3) 

If  the  phase  speed  Ci  is  a  function  of  frequency, 
which  it  will  be  if  ^'  is  not  constant,  the  trans- 
mission is  said  to  be  dispersive. 

2.    Experimental  Results 

A.    Insertion  Loss 

The  insertion  loss  for  a  4-mm  thick  sample  of 
diploe  as  a  function  of  frequency  is  shown  on 
figure  2.    These  data  show  very  little  loss  at  0.3 
MHz,  with  a  sharply  increasing  loss  between  0.7  and 
about  1.3  MHz.    At  higher  frequencies,  loss  in- 


-lol  1  1  1  1  1 

0  3  07  II  15  19 

FREQUENCY (MHz) 

Fig.  2.  Insertion  loss  of  a  4-rran  thick  sample  of 
diploe  compared  with  calculated  loss  due 
to  scattering  and  to  absorption  (A). 


creases  more  slowly  with  increasing  frequency, 
reaching  about  40  dB  at  2  MHz, 

B.    Phase  Speed 

The  insertion  phase  angle  for  a  4-mm  thick 
diploe  sample  is  shown  as  a  function  of  frequency 
on  figure  3.    We  see  that  the  phase  angle  does  not 


1000 

1  1 

1  1 

Cp  -  2800  m/s  »y  - 

800 

600 

400 

^^Cp-  2530  m/s 

200 

 Cp"  2190nv* 

1  1 

1  1 

ol  i  I  '  ^ 

"0  0  5  10  15  20 


FREQUENCY  (MHz) 

Fig.  3.    Insertion  phase  shift  of  a  4-mm  thick 
sample  of  diploe  showing  three  tangent 
regions  of  different  group  speed  Cg. 

increase  linearly  with  frequency.    Three  different 
values  of  phase  angle  are  noted  at  three  different 
frequencies.    The  phase-frequency  ratios  are 
used  together  with  eq.  (3)  to  calculate  the  phase 
speeds  at  the  three  different  frequencies.  The 
results  at  0.5  MHz  and  1.5  MHz  are  Ci  =  2190  m/s 
and  Ci  =  2530  m/s.    Similar  data  on  the  4-rim  thick 
diploe  sample  at  3.5  MHz  showed  c^  =  2800  m/s  [3]. 
This  tangent  is  sketched  on  figure  3,  where  it  is 
seen  to  match  with  the  data  for  frequencies  above 
2  MHz. 

C.    Scattered  Sound 

The  insertion-loss  measurement  apparatus  was 
modified,  as  follows,  to  measure  the  sound  scat- 
tered by  the  diploe  layer.    A  6-mm  diameter  core 
was  cut  from  the  skull  sample  and  positioned  co- 
axial ly  at  the  focus.    The  probe  was  then  scanned 
around  a  9-cm  radius  circle  centered  on  the  focus 
and  including  the  incident  sound  axis.    The  sound 
pressure  levels  at  9  cm  relative  to  the  sound 
pressure  level  at  the  focus  are  shown  on  figure  4 
as  a  function  of  observation  angle.    The  narrow 
beam  directed  back  towards  the  source  is  sound 
reflected  from  the  outer  table.    The  small  in- 
crease in  sound  pressure  level  at  180°  is  the 
transmitted  sound.    Sound  reaching  the  probe  at 
all  other  angles  has  been  scattered  by  the  skull 
core. 

3.    Discussion  of  Results 

A.    Insertion  Loss 

The  insertion  loss  is  shown  on  figure  2  to  be  a  ( 
rapidly  increasing  function  of  frequency.  The 
scattered  sound  shown  on  figure  4  is  approximately 
omnidirectional.    Both  of  these  experimental  facts 
lead  us  to  suspect  that  sound  is  being  scattered 
by  the  fat-  and  blood-filled  interstices  in  the 
spongy  bone  matrix.    Omnidirectional  scattering  is 
caused  by  small  discontinuities  in  the  bulk  modu- 


198 


ieo° 


Fig.  4.    Sound  pressure  levels  measured  9  cm  from 
a  core  of  diploe  caused  by  20-ps  pulses 
of  1-MHz  sound  incident  from  the  direction 
of  0  degrees. 

lus  of  the  medium.    If  the  average  bulk  modulus 
of  diploe  is  represented  by  Bq  and  the  bulk  modu- 
lus of  the  material  in  the  interstices  by  B,  we 
can  write  the  scattering  cross-sectional  area  a  of 
a  single  interstice  having  volume  V  [4]  as 

a  =  4^3X-^V2(B-B^/Bq)2    .  (4) 

We  consider  a  plane  wave  having  intensity  I  in- 
cident upon  a  layer  of  diploe  that  is  i  units 
thick  and  having  lateral  area  A.    The  total  power, 
Wg,  scattered  out  of  this  slice  having  volume  Aj, 
is  the  sum  of  power  scattered  by  all  N  interstices 
therein: 

N  N 
Ws  =  lE  CTj  =  I  4it3x-MaB/B^)2   ^  V2    .  (5) 

j=0  j 

We  simplify  the  expression  for  scattered  power  by 
observing  that  the  volume  of  the  slice  M  is  equal 
to  the  total  volume  of  scatterers  NV,  where  the 
average  volume  is 

N 

V  =  N-l   ^  Vj  . 
j 

We  define  also  the  normalized  variance  B  of  the 
interstitial  volumes. 

6  =  V2/V2  (6) 

Combining  eq.  (5)  with  eq.  (5),  upon  defining  the 
product  AI  as  the  power,  W^,  incident  upon  the 
slice,  we  have  the  following  result  for  the  ratio 
of  power  scattered  from  the  slice  to  power  incident 
upon  the  slice. 

W^/W.  =  4Tr3x-'*3  V  (aB/Bjj)25,  (7) 

Equation  7  is  equal  to  the  product  ai,  where  a 
is  the  intensity  attenuation  coefficient,  because 


the  power  balance  for  the  power,  W^-,  transmitted 
through  a  thin  slice  is  W^  =  Wi  -  Ws. 

W^/w.  =  1  -  W^/W.  =  e-"^  ^  1  -  al  (8) 

Dividing  by  incident  power      and  equating  to  the 
definition  of  attenuation  coefficient,  we  have  the 
desired  equality. 

The  average  bulk  modulus  of  the  diploe  sample 
was  calculated  from  measured  mass  density  and  sound 
speed.    The  result  was  Bq  =  pqCo^  =  7.81- IQi"  ybar. 
The  bulk  modulus  of  the  interstitial  contents  was 
taken  to  resemble  blood,  B  =  pc2  =  2.54-10io  ybar. 
The  average  interstitial  volume  was  calculated 
from  the  average  diameter  d  of  the  interstices, 
namely,  0.6  mm.    The  probability  density  distribu- 
tion for  interstice  volumes  is  not  known,  but  we 
will  assume  it  to  be  a  Rayleigh  distribution,  for 
which  B  =  4/tt. 

When  these  quantities  are  substituted  into  eq. 
(7),  we  have  for  the  attenuation  due  to  scattering 

ll/l  (dB/cm)  =  4.34a  =  13.2  F**  ,  (9) 

where  F  is  frequency  in  MHz. 

The  fundamental  result,  eq.  (4),  is  valid  only 
for  scattering  elements  that  have  diameters  smaller 
than  about  one-third  wavelength.    For  our  sample, 
then,  valid  results  occur  at  frequencies  less  than 
1.3  MHz.    At  higher  frequencies,  the  scattering 
law  saturates  and  changes  to  a  frequency-squared 
increase  [4].    Therefore,  the  complete  analytical 
representation  of  scattering  is  given  by  eq.  (10). 

Il/l  (dB/cm)  =  13.2  F**  F<F 

°  (10) 
=13.2  F2F2    F>F^  , 

where  Fo(MHz)  =  (ci/3d)  10"^. 

In  addition  to  attenuation  by  scattering,  there 
is  attenuation  by  absorption.    A  typical  value  of 
the  pressure  attenuation  coefficient  in  bone  is 
1.5  nepers/cm  at  1  MHz  [5].    The  associated  inser- 
tion loss  per  cm  is  ll/i  =  13. 2F.    The  total  at- 
tenuation, namely,  the  sum  of  absorption  and  scat- 
tering attenuation,  is  plotted  on  figure  2  for  a 
4-mm  thick  section.    The  agreement  is  very  good 
between  the  experimental  and  the  theoretical 
values  of  loss,  indicating  general  accuracy  of  the 
scattering  theory. 

B.    Sound  Power  Balance 

The  sound  power  balance  used  to  obtain  eq.  (10) 
can  be  checked  with  the  data  shown  on  figure  4. 
The  incident  sound  power,  W-j ,  is  given  in  terms  of 
the  incident  sound  Intensity,  I,  and  the  radius,  a, 
of  the  focus,  W-j  =  ira^  I.    The  reflected  sound 
power,  Wp,  is  given  in  terms  of  the  reflected  half- 
beamwidtn,  9,  and  the  reflected  inteREity,  Ip,  mea- 
sured at  distance  ro,  W^-  =  •iT(rosine)2  l„.  Tne 
scattered  sound  power,  Wg,  is  given  in  terms  of 
the  scattered  sound  intensity,  Ig,  measured  at 
distance  rg,      =  4TTr§  Ig. 

The  sound  power  balance  requires  power  incident, 
Wi ,  to  equal  the  power  scattered,  Wg,  plus  power 
reflected,  Wf,  plus  power  absorbed,  Wg.    We  ob- 
served in  the  preceding  section  that  attenuation  by 
absorption  is  equal  to  attenuation  by  scattering 
at  a  frequency  of  1  MHz.    The  scatter  data  on 
figure  4  are  for  this  frequency,  so  we  take  Wg  = 


199 


Wg.  Dividing  both  sides  of  the  power  balance  by 
the  incident  power,      ,  we  have  eq.  (11). 


Wi 


=  8{r^/a)Hl^/l.) 
+  (rpSin6/a)2(I^/I.; 


(11) 


The  value  of  Tq  is  9  cm,  the  value  of  e  is  about 
5°  (at  3-dB  down  point),  and  the  focal  radius  is 

2  mm.    The  value  of  Is/Ii  is  equal  to  10"'*-^  =  5,0- 
10"5  and  the  value  of  Ip/Ii  is  equal  to  10"^-^  = 
1.2-10"^,  because  the  reflected  pressure  is  19  dB 
less  than  the  incident  pressure  and  the  average 
scattered  pressure  is  43  dB  less  than  the  incident 
pressure.    Substitution  of  all  parameters  into 

eq.  (11)  gives  the  result  (W^+Wf+Wa)/!^!  =  0.994. 
Since  a  value  of  1.0  confirms  the  hypothesis,  this 
result  is  very  good,  and  the  sound  power  is  ac- 
curately accounted  for  by  reflection,  scattering, 
and  absorption.    The  transmitted  power  was  neglect- 
ed, since  it  is  scarcely  more  intense  than  the 
forward-scattered  sound. 

C.    Sound  Speed 

The  phase  speed,  Cp,  is  defined  in  terms  of  the 
wavenumber,  k:  Cp  =  lo/k.    This  is  the  speed  that 
an  observer  of  an  harmonic  wave  would  travel  in 
order  to  see  a  constant  wave  pressure.    The  group 
speed,  Cg,  is  defined  also  in  terms  of  the  wave- 
number:    Cg  =  dio/dk.    This  is  the  speed  that  an 
observer  of  a  "packet"  of  sound  energy,  having  by 
necessity  a  finite  bandwidth,  would  travel  in 
order  to  see  a  constant  wave  pressure.    The  two 
speeds  are  different  whenever  the  medium  is  dis- 
persive or  whenever  k  is  not  a  linear  function  of 
frequency.    We  see  clearly  from  the  data  on  figure 

3  that  sound  propagation  in  diploe  is  dispersive. 
This  result  is  consistent  with  our  finding  that 
attenuation  is  due  mainly  to  scattering,  for  sound 
propagation  is  then  dispersive  [6]. 

Experimental  measurements  of  sound  speed  in 
bone,  including  diploe,  have  almost  universally 
used  the  time-of-flight  of  a  sound  impulse  from 
which  to  calculate  sound  speed.    This  method 
yields  the  group  speed,  although  experimenters 
have  not  generally  noted  this.    Their  results  can- 
not properly  be  compared  with  measurements  of 
phase  speed  in  diploe,  unless  the  difference  is 
accounted  for. 

The  definition  of  group  speed  yields  eq.  (12), 
which  expresses  the  group  speed  in  terms  of  phase 
speed. 


Cp  df 


1- 1 


(12) 


The  phase  speed  calculation  from  our  data  on  the 
4~mm  skull  sample  (fig.  3  and  ref.  [3])  are  plotted 
on  figure  5.    We  see  that  a  simple  frequency-power 
law  fits  the  experimental  data  quite  well,  accord- 
ing to  Cp     f",  where  n  =  0.131.    In  this  case 
eq.  (12)  gives 


n)-i  =  1.15  c^ 


This  calculated  group  speed  is  plotted  on  figure  5 
together  with  the  experimental  values  of  phase 
speed. 


MEASURED  GROUP  SPEED,, 


CALCULATED  GROUP  SPEED  Cg 


MEASURED  PHASE  SPEED  Cp 


FREQUENCY  (MHz) 


Fig.  5. 


Measured  values  of  phase  speed  and  group 
speed  in  diploe  shown  with  group  speed 
calculated  from  the  phase  speed. 


The  sound  speed  of  the  4-mm  diploe  sample  was 
measured  by  the  time-of-flight  method,  using  three 
pulses  containing  most  of  their  energy  between  0.3 
to  0.4  MHz,  1.0  to  1.5  MHz,  and  2.0  to  3.0  MHz 
[3].    These  measurements  give  the  group  speed,  and 
the  values  are  plotted  also  in  figure  5.    It  can 
be  seen  that  the  measured  values  of  group  speed 
are  higher  than  measured  values  of  phase  speed, 
and  that  they  agree  rather  well  with  the  calculat- 
ed values. 

D.    Use  of  Focused  Beam 

A  focused  beam  was  used  to  make  the  measurements 
reported  herein  because  only  in  this  way  can  a  nar- 
row beam  be  generated  several  transducer-diameters 
away  from  the  transducer.    The  scattering  of  sound 
by  the  bone  adversely  affects  the  beam  shape  at  the 
focus  only  if  a  parameter  g  =  /tT      k^aL  exceeds 
unity  [7].    The  mean  square  deviation  of  the  index 
of  refraction  is  v^,  the  average  diameter  of  a 
scatterer  is  a,  and  the  distance  traversed  by  the 
sound  beam  is  L.    For  the  4  mm  thick  diploe  sample, 
the  value  of  the  parameter  g  is  only  about  0.01, 
so  that  no  sensible  beam  distortion  or  phase  aber- 
ration is  caused  at  the  focus  by  the  scattering  in 
the  diploe  layer. 

4.  Conclusions 

We  conclude  that  scattering  of  diagnostic  ultra- 
sound by  the  blood-  and  fat-filled  interstices  in 
diploe  dominates  sound  attenuation  at  frequencies 
above  about  0.7  MHz  and  also  introduces  dispersion. 

The  detrimental  effects  of  large  diploe  attenua- 
tion can  be  largely  avoided  by  using  interrogation 
pulses  that  contain  no  frequency  components  greater 
than  about  0.7  MHz.    Attenuation  due  to  scattering 
in  the  diploe  studied  equals  the  attenuation  due 
to  absorption  at  1  MHz  and  exceeds  it  at  higher 
frequences. 

The  detrimental  effects  due  to  diploe  disper- 
sion are  confined  only  to  the  higher  frequency 
range,  where  scattering  dominates  attenuation. 
Therefore,  if  low-frequency  pulses  are  used,  there 
will  be  little  dispersion.    If,  however,  high 
frequencies  are  necessary  to  obtain  trans-skull 
tissue  characteristics  by  spectral  analysis,  dis- 
persion can  significantly  reduce  range  resolution. 
For  example,  if  a  pulse  contained  energy  in  the 
band  from  0.5  to  2.0  MHz,  and  was  transmitted 
through  a  6-mm  thick  diploe  layer,  the  depth 


200 


smearing  in  the  brain  due  to  dispersion  would  be 
0.73  mm,  but  the  nominal  depth  resolution  of  the 
same  1.5-MHz  bandwidth  pulse  in  an  undispersive 
water- like  medium  is  0.5  mm. 

We  find  that  almost  all  reported  sound  speeds 
in  bone  have  been  obtained  by  the  time-of-flight 
pulse  method  and  are,  therefore,  actually  group 
speed.    There  is  no  evidence  that  sound  propaga- 
tion in  ivory  bone  is  dispersive,  so  these  data 
are  equivalent  to  phase  speed.    Since  we  find 
sound  propagation  in  diploe  to  be  dispersive,  most 
published  sound  speed  values  in  diploe  will  be 
higher  than  correct  values  for  phase  speed. 

References 

[1]    White,  D.  N.,  Clark,  J.  M. ,  Curry,  G.  R. ,  and 
Stevenson,  R.  J.,  The  Effects  of  the  Skull 
Upon  the  Spatial  and  Temporal  Distribution  of 
a  Generated  and  Reflected  Ultrasound  Beam 
(Ultramedison,  Kingston,  Ontario,  1976). 

[2]    Lizzi,  F.  L.    and  Laviola,  M.  A.,  Ultrasonic 
Spectral  Investigations  for  Tissue  Charac- 
terization in  Ultrasound  in  Medicine,  Denis 
White  and  Ralph  Barnes,  eds..  Vol.  11,  pp. 
427-39  (Plenum  Press,  New  York,  1976). 

[3]    Fry,  F.  J.    and  Barger,  J.  E.,  Acoustical 
properties  of  human  skull,  to  be  published 
in  J.  Acoust.  Soc.  Amer. 

[4]    Mason,  W.  P.    and  McSkimin,  H.  J.,  Attenua- 
tion and  scattering  of  high  frequency  sound 
waves  in  metals  and  glasses,  J.  Acoust.  Soc. 
Amer.  19,  472  (1947). 


[5]    Goldman,  D.  E.  and  Hueter,  T.  F. ,  Tabular 
data  of  the  velocity  and  absorption  of  high 
frequency  sound  in  mammalian  tissue,  J.  Acoust. 
Soc.  Amer.  28,  35-37  (1956). 

[6]    Howe,  M.  S.,  Wave  propagation  in  random  media, 
J.  Fluid  Mech.  45,  769-83  (1971). 

[7]    Chernov,  L.  A.,  Wave  Propagation  in  a  Random 
Medium  (Dover  Publications,  Inc.,  New  York, 
1967). 


201 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


TRANSKULL  TRANSMISSION  OF  AXISYMMETRIC  FOCUSED  ULTRASONIC  BEAfIS 
IN  THE  0.5  TO  1  MHZ  FREQUENCY  RANGE: 
IMPLICATIONS  FOR  BRAIN  TISSUE  VISUALIZATION,  INTERROGATION,  AND  THERAPY 


F.  J.  Fry 

Ultrasound  Research  Laboratories  of  the 
Indianapolis  Center  for  Advanced  Research 
Indianapolis,  Indiana    46202,  U.S.A. 


In  order  for  ultrasound  to  become  a  practical  clinical  technique  for  diagnosis  of 
many  intracerebral  diseases  in  the  adult  human,  it  has  been  necessary  to  answer  ques- 
tions concerning  insertion  loss  of  skull,  temporal  and  spatial  characteristics  of 
beams  operating  in  the  0.5  to  1.0  MHz  range  (the  requisite  frequency  is  a  function 
of  the  particular  adult  skull)  have  a  maximum  single  pass  skull  insertion  loss  of 
nearly  10  dB  (20  dB  for  a  pulse  echo  system)  which  can  be  handled  with  present  tech- 
niques to  provide  adequate  signal  strength  from  normal  and  pathological  features  of 
brain.    In  this  frequency  range,  for  skulls  in  our  studies,  the  appropriately  se- 
lected frequency  is  unchanged  after  double  pass  skull  transmission,  the  6  dB  beam 
width  is  increased  by  a  maximum  of  40  percent  and  the  beam  focus  is  shifted  laterally 
by  a  maximum  of  3  mm.    Resolution  of  string  targets  or  live  brain  targets  has  been 
demonstrated  to  be  in  the  2  to  3  mm  range  at  1  MHz  and  at  0.5  MHz  it  appears  that 
4  to  6  mm  resolution  can  be  achieved. 

A  high  intensity  focused  ultrasonic  beam  (1  MHz)  has  been  transmitted  through  an 
excised  adult  skull  and  used  to  produce  a  focal  thermal  flaw  in  lucite.    This  simu- 
lation test  indicates  that  the  induction  of  transkull  focal  lesions  in  live  adult 
brain  may  now  be  possible. 

Keywords:    Axisymmetric;  beams;  focal  thermal  flaw;  skull  transmission;  ultrasound. 


1.  Introduction 

In  the  adult  human  skull  the  diploe  layer  is  a 
dominant  factor  in  determining  the  magnitude  of 
insertion  loss  for  a  transmitted  ultrasonic  beam. 
Insertion  loss  characteristics  for  human  skull  as 
a  function  of  frequency  have  been  documented  in 
the  literature  [1-4]^  but  only  recently  has  the 
acoustic  character  of  the  specific  skull  layers 
been  more  clearly  defined  [5].    These  measure- 
ments indicate  that  frequencies  less  than  0.5 
j     MHz  are  necessary  to  have  a  single  pass  insertion 
I     loss  less  than  10  dB  for  some  adult  human  skulls, 
i     Skulls  we  have  studied  with  a  loss  of  this  magni- 
i|     tude  at  a  given  frequency  will  transmit  an  ultra- 
I     sonic  signal  of  this  same  frequency  with  minimal 
jj     temporal  distortion.    For  infants,  the  skull  in- 
f     sertion  loss  at  ordinarily  used  diagnostic  fre- 
quencies (2.25  MHz)  is  less  than  20  dB  since 
there  is  essentially  no  diploe  layer.    Since  the 
two-way  insertion  loss  is  twice  the  one-way  loss 
in  a  pulse  echo  regime,  the  20  dB  maximum  skull 
insertion  loss  (at  the  appropriate  frequency)  is 
not  a  serious  impediment  to  pulse  echo  visual i- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


zation  and  interrogation  of  brain  features. 
Preservation  of  temporal  coherence  in  the  wave 
is  also  an  important  feature  and  this  can  be 
achieved  when  the  appropriate  ultrasonic  fre- 
quency range  is  selected. 

Use  of  an  axisymmetric  focused  beam  in  pre- 
vious studies  indicated  that  preservation  of  the 
beam  spatial  focal  character  was  achieved  in  a 
qualitative  sense  and  that  the  maximum  lateral 
displacement  of  the  beam  was  of  the  order  of 
3  mm.    Studies  with  phased  array  systems  indi- 
cated a  need  for  phase  compensation  at  the  re- 
ceiver array  to  correct  for  local  skull  thick- 
ness variations  which  change  the  arrival  time  of 
the  waves  at  the  receiver  elements  [6-8].  Such 
variations  also  lead  to  variable  insertion 
1 osses  over  the  skul 1  . 

This  report  covers  studies  in  which  low  in- 
tensity axisymmetric  focused  ultrasonic  beams 
suitable  for  diagnostic  medicine  have  been 
transmitted  through  excised  adult  human  skull 
sections.    The  beams  have  been  studied  for  their 
temporal  and  spatial  preservation  as  well  as 
displacement  after  skull  transmission. 
Transkull  B-scan  presentations  of  physical 
targets  and  live  human  brain  indicate  consid- 
erable potential  for  clinical  diagnostic 
medicine. 


203 


No  previous  literature  seems  to  exist  which 
indicates  the  possibility  of  transmitting  an 
intense  focused  ultrasound  beam  through  an  adult 
human  skull  which  would  be  capable  of  producing 
focal  lesions  in  brain.    A  series  of  experiments 
reported  here  shows  that  a  high  intensity  fo- 
cused ultrasonic  beam  of  a  frequency  appro- 
priate for  a  given  skull  can  be  transmitted 
through  the  excised  adult  human  skull  and  pro- 
duce a  thermal  focal  "lesion"  in  lucite  [9]. 
Previous  studies  show  that  lucite,  in  the 
physical  model  sense  used  here,  can  simulate 
1 ive  brain  tissue  [10] . 

2.    Materials  and  Methods 

Measurement  of  frequency  dependent  attenua- 
tion loss  in  skull  was  made  with  a  system  de- 
scribed elsewhere  [5].    A  schematic  diagram  of 
the  essential  components  is  shown  in  figure  1. 


Fig.  1.    Schematic  diagram  for  skull  insertion  loss 
measurements  as  a  function  of  frequency. 


PA  = 

pul ser 

T  = 

focused  transducer 

S  = 

skull  (formalin  fixed) 

P  = 

probe 

D  = 

digitizer  (8  bit  100  MHz  Biomation 

8100) 

C  = 

computer  (PDP-11/45) 

PP  = 

printer/ploter  (Versatec  1100) 

x-y-z  = 

(3  orthogonal)  motion  coordinate 

system 

Sound  from  the  transducer  passes  through  the 
skull  section  under  study  and  is  received  on  a 
3  mm  diameter  ferroelectric  probe  (10  MHz  fun- 
damental resonance).    The  acoustic  signal  from 
the  probe  is  analyzed  by  an  FFT  (Fast  Fourier 
Transform)  method  in  the  skull  and  non-skull 
case,  and  the  difference  of  the  Fourier  coeffi- 
cients at  each  frequency  is  taken  as  the  inser- 
tion loss  at  that  frequency. 

Measurements  of  beam  configuration  and  dis- 
placement after  skull  transmission  were  made 
by  moving  the  probe  in  a  direction  perpendicu- 
lar to  the  sound  transmission  axis.    The  skull 
is  also  mounted  on  a  coordinate  system  so  that 
it  can  be  moved  in  3  orthogonal  directions. 
Rotational  axes  are  also  provided  so  that  the 
central  ray  of  the  transducer  can  be  angulated 
with  respect  to  the  skull  surface. 

B-scans  of  physical  models  or  patients  were 
made  with  the  apparatus  shown  schematically  in 
figure  2.    This  apparatus  provides  a  mechani- 


Fig.  2.    Schematic  diagram  for  B-scan  imaging. 
PA  =  patient 
T  =  tranceiver 
W  =  water  bath 

X  =  mechanically  driven  linear  motion 
P  =  pul ser 
A  =  ampl if ier 
D  =  detector 
SC  =  scan  converter 
M  =  monitor 

cally  driven  linear  scan  path.    Such  a  path  is 
very  limiting  for  scans  involving  skull  since 
it  is  important  to  approximately  follow  the 
skull  contour  so  that  the  central  transducer 
ray  is  normal  to  the  skull  surface  for  minimal 
insertion  loss.    In  order  to  have  minimal  beam 
displacement,  this  normal  angulation  configu- 
ration should  be  preserved. 

For  the  high  intensity  focused  beam  study 
involving  production  of  thermal  "lesions"  in 
lucite  (lucite  used  here  as  a  brain  simulation 
target)  the  system  configuration  is  shown  in 
figure  3.  A  1  MHz  high  power  transducer  [11] 
with  a  6  dB  down  (ultrasound  intensity)  beam 
diameter  of  2.2  mm  was  used  for  this  part  of 
the  study.    The  skulls  used  were  obtained  from 


Fig.  3. 


Schematic  diagram  for  high  power 
transkull  test. 

T  =  high  power  focused  transducer 
S  =  skull  (formalin  fixed) 
L  =  lucite  block  (2  inch  x  2  inch  x 
2  inch) 

W  =  temperature-controlled  water  bath 
PA  =  power  amplifier  (1  kW) 
TG  =  timing  gate 
FS  =  frequency  source 


204 


patients  at  autopsy  and  were  immediately  placed 
in  Ringer's  solution  and  stored  in  the  refrig- 
erator.   By  the  time  the  experiments  reported 
here  were  conducted,  the  skulls  had  been  placed 
in  formalin  and  stored  at  room  temperature. 
The  storage  procedure  has  been  shown  to  provide 
the  necessary  conditions  for  long-term  stabil- 
ity of  the  acoustical  properties  of  skull 
starting  from  the  fresh  excised  state  proper- 
ties [12]. 

3.  Results 

One  particular  adult  human  skull  has  been 
selected  from  a  group  used  in  the  preparation  of 
an  extensive  manuscript  [5]  on  acoustic  properties 
of  skull  to  illustrate  the  frequency  filtering 
properties  of  adult  skull  having  a  diploe  layer. 
Figure  4  shows  the  free  field  probe  received 
waveform  transmitted  from  a  focused  transducer 
with  a  damped  resonant  frequency  of  2.5  MHz. 
When  the  skull  is  inserted  between  the  trans- 
ducer and  the  probe  the  signal  strength  is 
greatly  attenuated,  as  will  be  described  later. 


A 


•|  0. 5  ys/division 


Fig.  4.    Probe  received  waveforms  (using  setup 
shown  on  fig.  1  with  an  oscilloscope 
connected  in  fromt  of  D). 
B  =  without  skull  (2.5  MHz) 
A  =  with  intervening  skull  (skull  filter- 
ing permits  0.56  MHz  transmission) 


and  the  dominant  transmitted  frequency  is  0.56 
MHz.    When  a  0.5  MHz  frequency  damped  pulse  is 
transmitted  through  the  skull,  reflected  from 
a  glass  plate  and  passed  again  through  the 
skull,  the  received  waveform  recorded  from  the 
original  transducer  is  shown  on  figure  5.  The 
double  pass  transmission  loss  is  10  dB  for  this 
skull.    Use  of  ultrasonic  frequencies  in  this 
range  (0.5  MHz)  is  an  advantage  because  of  the 
temporal  coherence  preservation. 

Spatial  aspects  of  beam  displacement  and 
distortion  after  skull  transmission  for  a  0.5 
MHz  focused  beam  is  demonstrated  on  figure  6. 
These  beam  aspects  have  been  studied  with  skull- 
to-probe  distances  from  2  to  10  cm,  which  is  the 
range  needed  for  brain  target  delineation  in  the 
intact  skull  case  with  the  live  patient.  In 
these  two  adult  skulls  the  maximum  beam  dis- 


A 


i  0.5  ys/division 


Fig.  5.    Transceiver  waveforms  after  reflection 
from  glass  plate  (using  setup  shown  on 
fig.  1  with  an  oscilloscope  connected 
in  front  of  D) . 

A  =  without  skull  (0.5  MHz,  distortion 
is  due  to  the  electrical  driving 
network) 

B  =  after  skull  transmission  (0.5  MHz, 
skull  has  filtered  out  the  high 
frequency  components  in  the  A 
waveform  but  the  0.5  MHz  fundamental 
has  been  preserved. 


205 


■100 
■  90 
■80 
•70 
1-60 
50 
1-40 
30 


500  kHz  tranducer 
4  mm  diameter  disc  probe 

free  field  plot 
L  =  2  cm  (skull  in  place) 
L  =  4  cm 
L  =  6  cm 
L  =  8  cm 
F  =  L  =  10  cm 


Fig.  6.    Beam  distortion  and  displacement  after  single  pass  skull  transmission  as  a  function  of  probe- 
to-skull  distance  (L)  for  a  fixed  transducer-to-probe  distance  (20  cm).    Lateral  beam  plots 
and  beam  displacement  obtained  by  moving  the  probe  for  the  field  plots  (using  setup  shown  on 
fig.  1  with  oscilloscope  connected  in  front  of  D  to  display  the  receiving  probe  output  as  a 
function  of  space  coordinate).    Insertion  loss  variations  for  skulls  6  and  8  were  within 
±  0.1  dB  of  the  mean  insertion  loss  value  for  all  skull-to-probe  distances  L.    Curves  for 
each  L  distance  are  shown  vertically  displaced  on  the  graph  for  reading  convenience. 


placement  from  the  free  field  condition  is  2.5 
mm  for  any  probe-to-skull  position,  and  the 
maximum  insertion  loss  variations  for  the  same 
range  were  within  ±  0.1  dB  of  the  mean  value  for 
all  axial  distances  studied.    These  studies  also 
show  that  the  6  dB  beam  width  is  increased  by  a 
maximum  of  40  percent.    Other  studies  show  [5] 
that  when  a  focused  transceiver  is  used  with  the 
probe  as  a  target,  the  6  dB  beam  widths  de- 
crease in  all  cases  over  the  6  dB  beam  widths 
in  the  single  pass  transmission  case  with  the 
probe  as  receiver. 

On  the  basis  of  temporal  and  spatial  coher- 
ence and  beam  displacement,  a  0.5  MHz  frequency 
is  appropriate  for  adult  human  skulls.    The  in- 
sertion loss  characteristic  as  a  function  of 
frequency  for  adult  human  skull,  shown  on  figure 
7,  indicates  the  desirability  of  this  frequency 
(0.5  MHz)  from  a  system  sensitivity  viewpoint  if 
skull  insertion  loss  is  to  be  minimized.  Given 
these  indicators  for  using  0.5  MHz  ultrasonic 
frequency  for  transkull  visualization  and  inter- 
rogation of  brain,  the  important  remaining  aspect 
is  resolution.    Within  the  scope  of  these  studies, 
using  a  very  preliminary  transceiver  prototype 
(0.5  MHz)  and  no  signal  processing  to  provide 
image  enhancement,  the  scan  of  a  string  target 
without  intervening  skull  is  shown  on  figure  8. 
Although  the  image  quality  is  not  ideal,  it  can 
be  seen  that  targets  spaced  5  mm  apart  in  real 
space  can  be  resolved.    Note  also  that  range 
resolution  has  not  been  optimized  since  this 
prototype  transceiver  is  not  highly  damped. 


+-> 
+-> 


Fig.  7. 


28- 


24 


20 


16 


12 


0.6  1.0 
Frequency,  MHz 


1.4 


1 .8 


Insertion  loss  as  a  function  of  frequency 
for  single  pass  skull  transmission  (skull 
6  section  -  formalin  fixed). 


206 


Skull 


Fig.  8.    Monofilament  nylon  string  target  (0.01 
inch  string  diameter)  with  0.5  MHz 
focused  transducer  visualized  with 
system  shown  on  fig.  2. 
A  =  without  skull  lowest  set  of  3  x  5 

array  has  strings  1  cm  apart  on 

centers . 
B  =  with  skull 

This  resolution  will  be  improved  with  a  larger 
aperture  angle,  and  with  appropriate  signal 
processing  this  resolution  will  be  apparent 
over  a  wide  dynamic  range  of  signal  strengths. 
When  the  skull  is  interposed  there  are  targets 
which  show  evidence  of  lateral  resolution  simi- 
lar to  that  in  the  non-skull  case.    Note  that 
because  of  this  scanning  system's  inability  to 
follow  the  skull  contour  not  all  strings  can  be 
visualized.    A  scan  mode  in  which  the  scanning 
transducer  axis  is  maintained  perpendicular  to 
the  skull  surface  at  the  point  of  passage  of 
the  central  ray  of  the  focused  beam  will  resolve 
this  problem. 

Our  studies  include  a  transkull  visualization 
ofa  glioblastoma  in  a  live  adult  patient  in 
which  an  x-ray  transaxial  tomographic  scan  was 
also  made.    Results  of  these  scans  are  shown  on 
figure  9.    In  this  case  the  skull  insertion  loss 
was  small  enough  so  that  a  1  MHz  focused  trans- 
ceiver could  be  used  (this  unit  is  somewhat  more 
optimal  than  the  0.5  MHz  prototype).    The  tumor 
tissue  was  verified  by  pathological  examination 
at  the  time  of  surgery.    The  computerized  axial 


lateral 
ventricles 


presumed 

tumor 

region 


EMI  scan,  live  patient 


focus 
marker 


lateral 
ventricl e 


hard 
gristle 

surgical ly 
excised 
patholog- 
ically 
verified 
tumor 

tissue  (1 ) 


Fig.  9.    B-scan  of  live  human  patient  brain  tumor 
compared  to  CAT  (computerized  axial 
tomography)  scan.    These  scans  are  of  a 
horizontal  section  through  the  brain. 
The  ultrasound  scans  were  made  with  the 
system  shown  on  fig.  2. 
A  =  CAT  scan 
B  =  ultrasonic  scan 

tomographic  scan  showed  presumed  evidence  of  a 
tumor  on  normal  brain  structural  grounds  (ven- 
tricular outlines  and  displacement),  but  not  on 
a  tissue  density  difference  basis.    The  ultra- 
sound scan  clearly  shows  tissue  differentiation. 
The  linear  motion  scan  mode  used  does  not  follow 
the  skull  contour  so  that  the  patient-transceiver 
angle  was  selected  to  permit  best  viewing  over 
the  tumor  region.    Spatial  resolution  of  some 
targets  in  the  lateral  direction  is  2  mm  and  in 
the  longitudinal  direction  is  3  mm.    It  is  an- 
ticipated that  resolution  using  an  appropriate 
0.5  MHz  transceiver  will  be  no  worse  than  twice 
these  dimensions. 

The  above  material  is  relevant  to  ultrasonic 
regimes  for  transkull  diagnostic  purposes  and 
as  such  uses  ultrasonic  intensities  appropriate 
to  that  usage  (low  milliwatt  average  intensity 
range).    The  subsequent  material  covers  experi- 
ments demonstrating  transmission  through  excised 
adult  human  skull  of  intense  ultrasonic  focal 
beams  suitable  for  such  interactions  as  focal 
lesion  production  in  brain. 

For  this  study,  the  high  power  transducer  was 
used  to  produce  focal  thermal  "lesions"  in  the 
lucite  block  brain  tissue  simulator  [10].  These 
lesions  were  first  produced  with  no  skull  inter- 
vening in  the  acoustic  path.    The  human  skull 
section  was  then  inserted  in  the  acoustic  path 


207 


and  focal  lesions  were  produced  of  a  size  simi- 
lar to  those  produced  in  the  non-skull  case. 
A  number  of  focal  thermal  lesions  were  produced 
at  various  distances  from  the  lucite  surface 
(typically  2  cm  deep).    The  skull  intervening 
beam  plot  compared  to  the  free  field  case  is 
shown  on  figure  10.    The  single  pass  insertion 
loss  was  12  dB  for  this  skull  at  the  1  MHz 
operating  frequency  of  the  transducer.  More 
extensive  coverage  of  this  work  is  provided  in 
other  material  [9] . 


2  10  1 

Distance,  mm 

Fig.  10.    Lateral  beam  plot  obtained  with  high 
intensity  focused  ultrasonic  trans- 
ducer, with  and  without  intervening 
skull.    Measurements  made  with  systems 
shown  on  fig.  1 . 
A  =  without  skul 1 
B  =  with  skull,  the  beam  axis  was 
shifted  2  mm  laterally  from  the 
free  field  position  (A)  and  the 
insertion  loss  was  12  dB  so  no 
direct  intensity  comparisons 
should  be  made  from  this  figure. 

4.  Conclusion 

This  work  has  shown  that  for  ultrasound  to 
have  a  significant  impact  in  medicine  for  diag- 
nosis and/or  therapeutics  of  brain  disorders,  it 
is  necessary  to  use  ultrasonic  frequencies  in  the 
0.5  to  1.0  MHz  range.    In  this  range,  the  single 
pass  insertion  losses  can  be  held  to  a  maximum 
of  10  to  12  dB  and  the  temporal  coherence  of  the 
wave  can  be  maintained.    It  has  also  been  shown 
that  the  focused  beam  configuration  can  be  main- 
tained after  skull  transmission  so  that  the  ad- 
vantages of  resolution  and  intensity  gain  can  be 
utilized.    In  diagnosis,  these  advantages  lead 
to  excellent  pictorial  images  of  brain  struc- 
tures and  ability  to  perform  tissue  interrogation 
on  temporarily  undisturbed  waves  due  to  skull 
transmission.    For  potential  therapeutics  in 
brain,  these  ultrasonic  techniques  offer  the  pos- 
sibility of  transkull  ablation  of  tumor  tissue 
or  other  tissues  such  as  that  involved  for  focal 
epilepsy  and  modification  of  blood-brain  barrier 
for  enhancing  chemotherapeutics  of  tumor  therapy 
or  other  brain  disorders. 

When  appropriate  ultrasonic  frequencies  and 
focal  beams  are  used,  it  seems  apparent  that  a 
significant  impact  can  now  be  made  in  brain  diag- 
nosis and  possibly  therapeutics  with  ultrasonic 
devices  and  techniques. 


Acknowledgments 

Work  reported  here  was  supported  by  NIH  Grant 
No.  NOl-NS-3-2319,  NSF  Grant  No.  APR75-14487  and 
The  Indianapolis  Center  for  Advanced  Research. 

References 

[I]  Heuter,  T.  F.,  Cavlieri,  A.,  Langmuir,  B., 
Butkus,  W.,  Kyrazia,  D.,  Ballantine,  H.  T.  and 
Bolt,  R.  H.,  The  Detection  of  Intracranial 
Tumors  by  Use  of  Ultrasound,  Quarterly  Progress 
Report,  Acoustics  Laboratory,  Massachusetts 
Institute  of  Technology,  Cambridge,  Mass., 

pp.  38-46,  July/September  (1951). 

[2]    Goldman,  D.  E.  and  Hueter,  T.  F.,  Tublar  data 
of  the  velocity  and  absorption  of  high  fre- 
quency sound  in  mammalian  tissue,  J.  Acoust. 
Soc.  Amer.  28  (1),  35-37  (1956). 

[3]    Martin,  B.  and  McElhaney,  J.  F.,  The  Acoustic 
Properties  of  Human  Skull  Bones,  in  Biomedical 
Material  Research,  Vol.  5,  pp.  325-333  (John~ 
Wiley  &  Sons,  Inc.,  New  York,  1971). 

[4]    Smith,  J.  W.,  Phillips,  D.  J.,  von  Ramm,  0.  T. 
and  Thurstone,  F.  L.,  Real  Time  B-Mode  Echo- 
Encephalography,  in  Ultrasound  in  Medicine, 
Vol.  2,  D.  White,  ed.,  pp.  373-381  (Plenum 
Press,  New  York,  1976). 

[5]    Fry,  F.  J.  and  Barger,  J.  E.,  Acoustical 
Properties  of  Human  Skull  (submitted  to 
J.  Acoust.  Soc.  Amer. ,  1977). 

[6]    Phillips,  D.  J.,  Smith,  S.  W. ,  von  Ramm,  0.  T. 
and  Thurstone,  F.  L.,  A  Phase  Compensation 
Technique  for  B-Mode  Echo-encephalography ,  in 
Ultrasound  in  Medicine,  D.  White,  ed..  Vol.  1, 
pp.  345-404  (Plenum  Press,  New  York,  1975). 

[7]    Smith,  S.  W.,  Mitter,  E.  B.,  von  Ramm,  0.  T., 
and  Thurstone,  F.  L.,  Signal  Processing  Tech- 
niques for  Improving  B-Mode  Echo-encephalog- 
raphy, in  Ultrasound  in  Medicine,  D.  White,  ed. 
Vol.  1,  pp.  405-414  (Plenum  Press,  New  York, 
1975). 

[8]    Phillips,  D.  J.,  Smith,  W.  W. ,  von  Ramm,  0.  T., 
and  Thurstone,  F.  L.,  Sampled  Aperture  Tech- 
niques Applied  to  B-Mode  Echo-encephalography, 
in  Acoustical  Holography,  N.  Booth,  ed..  Vol. 6, 
pp.  103-120  (Plenum  Press,  New  York,  1975). 

[9]    Fry,  F.  J.,  Transkull  Transmission  of  an  In- 
tense Focused  Ultrasonic  Beam,  Ultrasound  Med. 
Biol  ■  3  (2,3),  179-184  (1977). 

[10]  Lele,  P.  P.,  Irradiation  of  plastics  with 

focused  ultrasound:    a  simple  method  for  eval- 
uation of  dosage  factors  in  neurological  appli- 
cations, J.  Acoust.  Soc.  Amer. ,  34  (4),  412- 
420  (1962y~  ~ 

[II]  Fry,  F.  J.,  Heimburger,  R.  F.,  Gibbons,  L.  V., 
and  Eggleton,  R.  C,  Ultrasound  for  Visualiza- 
tion and  Modification  of  Brain  Tissue,  I . E . E. E. 
Trans.  Sonics  and  Ultrasonics  SU-17  (3),  165- 
169  (1970y:    ~  ~~ 

[12]  White,  D.  N.,  Clark,  J.  M.,  Curry,  G.  R.,  and 
Stevenson,  R.  J.,  The  Effects  of  the  Skull  Upon 
the  Spatial  and  Temporal  Distribution  of  a 
Generated  and  Reflected  Ultrasonic  Beam,  avail- 
able from  Ul tramedi son.  Box  763,  Kingston, 
Ontario,  Canada  (1976). 


208 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525  (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


SOME  ADVANCES  IN  ACOUSTIC  IMAGING  THROUGH  SKULL 


S.  W.  Smith 

Food  and  Drug  Administration 
Rockville,  Maryland    20852,  U.S.A. 

D.  J.  Phillips 

Center  for  Biomedical  Engineering 

University  of  Washington 
Seattle,  Washington    98105,  U.S.A. 

0.  T-  von  Ramm  and  F.  L.  Thurstone 

Department  of  Biomedical  Engineering 

Duke  University 
Durham,  North  Carolina    27706,  U.S.A. 


Previous  attempts  to  image  the  adult  brain  through  the  skull  using  diagnostic 
ultrasound  have  resulted  in  images  of  poor  lateral  resolution  and  limited  dynamic 
range..    The  skull  can  be  modeled  as  an  acoustic  lens  whose  attenuation  increases 
rapidly  above  1  MHz  and  whose  thickness  variations  introduce  phase  aberrations  on 
the  order  of  several  wavelengths  across  the  transducer  aperture.    Statistical  analy- 
sis of  skull  thickness  data  indicate  that  an  electronic  sector  scanner  using  a  1  MHz 
linear  array  transducer  is  less  sensitive  to  these  effects  of  the  skull  than  tradi- 
tional pulse  echo  systems  operating  at  higher  frequencies.    Representative  ultra- 
sound tomograms  of  the  brain  are  shown.    In  addition,  water  tank  experiments  are  de- 
scribed in  which  the  skull  phase  aberration,  ^{x),  was  measured  on  line  and  removed 
by  incorporating  a  compensating  phase  variation,  -<t>(x),  into  the  transmit  and  re- 
ceive timing  of  a  digitally  controlled  real  time  phased  array  imaging  system.  Pre- 
liminary results  show  that  the  lateral  resolution  of  the  imaging  system  is  restored 
in  both  the  transmit  and  receive  modes. 

Key  words:    Brain  scanning;  echoencephalography ;  neurology;  phase  compensation. 


1,  Introduction 

Since  the  initial  development  of  B-mode  ultra- 
sonography, there  have  been  many  attempts  (Fry  et_ 
al .  [1]^;  Brinker  and  Taveras  [2])  to  produce 
TiTgh  quality  cross-sectional  images  of  the  brain 
through  the  intact  adult  skull.    Real  time  B-mode 
echoencephalography  was  initially  demonstrated  by 
Somer  in  1968  using  phased  array  techniques.  De- 
spite these  efforts,  B-mode  echoencephalography 
is  seldom  used  clinically  because  of  the  poor 
quality  of  the  images  obtained  with  current  tech- 
niques.   One  reason  for  the  immediate  widespread 
application  of  computerized  tomography  in  the 
head  has  been  the  failure  of  B-mode  ultrasonic 
imaging  of  the  brain.    The  intervening  presence 
,   of  the  skull  results  in  poor  signal  to  noise 
ratio,  limited  dynamic  range  and  reduced  lateral 
resolution. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


Three  characteristics  of  skull  bone  are  re- 
sponsible for  poor  quality  brain  images:  (1) 
substantial  attenuation  of  diagnostic  ultrasound 
by  the  skull  bone;  (2)  rapid  increase  of  that  at- 
tenuation with  the  frequency  of  ultrasound;  and 
(3)  the  variation  of  thickness  of  the  inner  table 
of  the  skull.    In  order  to  optimize  B-mode  ultra- 
sound images  of  the  brain,  these  aberrating  ef- 
fects of  the  skull  bone  must  be  overcome.    By  a 
proper  choice  of  transducer  frequency,  aperture 
size,  image  format,  and  the  use  of  effective 
signal  processing  techniques,  significant  im- 
provement may  be  achieved  in  ultrasound  tomograms 
of  the  brain.    In  this  manuscript  these  three 
problems  of  acoustic  imaging  through  the  skull 
will  be  examined.    In  addition  one  promising 
technique  will  be  described  in  which  the  effects 
of  phase  variations  due  to  the  skull  have  been 
removed  by  a  phase  compensation  process  imple- 
mented with  the  real  time,  swept  focus,  phased 
array  imaging  system  under  current  evaluation  at 
Duke  University  (von  Ramm  and  Thurstone  [3]). 


209 


2.    Ultrasound  Attenuation  of  the  Skull 

A  propagating  ultrasonic  pulse  is  strongly  at- 
tenuated after  two  passages  through  the  skull  by 
such  phenomena  as  diffractive  scattering,  mode 
conversion,  absorption,  and  the  impedance  mis- 
match at  the  skull-brain  interface  (White  [4]). 
Previous  measurements  (Hueter  [5];  Smith  et  al . 
[11];  Fry  et  al .  [6])  indicate  that  the  attenua- 
tion of  ultrasound  by  the  skull  is  strongly  fre- 
quency dependent  increasing  from  approximately 
10  to  20  dB/cm  at  1  MHz  to  50  to  60  dB/cm  at 
2  MHz. 

In  figure  1  the  solid  curve  shows  skull  at- 
tenuation data  from  Hueter  [5]  normalized  to  a 

1  cm  thickness.    The  dashed  line  shows  similar 
data  which  we  obtained  from  a  preserved  segment 
of  the  adult  skull.    The  data  indicate  that  a 

2  MHz  signal  is  attenuated  approximately  50  dB/cm 
of  travel  through  skull  bone.    The  attenuation  at 
1  MHz  is  approximately  12  dB/cm.    Much  of  the 
energy  lost  from  the  interrogating  pulse  returns 
to  the  imaging  system  as  unwanted  acoustic  noise. 


FREQUENCY  (MHz) 


Fig.  1.    Attenuation  of  skull  bone  versus 
frequency. 


Such  a  high  background  reduces  the  signal  to 
noise  ratio,  obscuring  low  level  echo  information 
and  restricting  the  useful  dynamic  range  of  the 
imaging  system.    The  result  is  usually  a  high 
contrast  B-mode  image  of  the  brain  in  which  only 
the  strong  specular  echoes  are  displayed.  From 
the  figure  it  can  be  seen  that  using  a  1  MHz 
transducer  substantially  reduces  the  problem  of 
high  attenuation  due  to  the  skull. 

The  rapid  increase  of  attenuation  with  fre- 
quency also  degrades  the  lateral  resolution  of  a 
B-mode  imaging  system  (Smith  et  al .  [11]).  For 
a  typical  broadband  diagnostic  transducer,  the 
ultrasonic  pulse  contains  significant  amounts  of 
energy  above  and  below  the  center-frequency. 
After  two  passes  through  an  adult  skull  the  high 
frequency  content  of  the  pulse  has  been  signifi- 
cantly reduced  relative  to  the  low  frequency 
components.    The  center  frequency  of  the  ultra- 
sound is  effectively  shifted  to  a  lower  fre- 
quency.   For  a  fixed  transducer  aperture,  such  a 
frequency  shift  will  result  in  degraded  lateral 
resolution.    Our  experience  with  linear  array 
transducers  and  broadband  piston  transducers 
has  indicated  that  an  interrogating  2  MHz  pulse 
will  show  a  center  frequency  of  approximately 
.8  MHz  after  two  passes  through  the  temporal 
region  of  an  adult  skull.    This  results  in  more 
than  a  twofold  loss  in  resolution  capability  in 
the  lateral  dimensions.    A  1  MHz  ultrasonic  pulse 
is  also  shifted  to  approximately  .8  MHz  but  the 
resultant  loss  of  resolution  is  not  as  signifi- 
cant since  the  original  transducer  aperture  will 
normally  be  larger  at  1  MHz.    For  acoustic  imag- 
ing through  the  skull,  a  1  MHz  transducer  would 
appear  to  be  a  reasonable  compromise  considering 
the  requirements  of  system  resolution,  adequate 
sensitivity  and  transducer  size. 

3.    Thickness  Variation  of  the  Skull  Bone 

Of  equal  importance  is  a  consideration  of  the 
effects  of  the  thickness  variation  of  the  skull 
bone  on  image  quality.    In  the  temporal  and 
parietal  areas  of  the  skull  where  most  neuro- 
logical ultrasound  examinations  are  made,  the 
inner  table  of  the  skull  bone  undergoes  variation 
in  thickness  on  the  order  of  1  to  2  mm.    It  has 
been  shown  that  phase  variations  are  introduced 
across  the  transducer  aperture  when  acoustic 
energy  propagates  through  such  a  section  of  skull 
(Phillips  et  al.  [8]).    Figure  2  (Phillips  et  al . 
[8])  shows  two  elements  of  a  one  dimensional 
transducer  array  simultaneously  transmitting  an 
acoustic  pulse  through  a  section  of  skull.  (An 
analogous  situation  would  apply  in  the  receive 
mode.)    Since  the  acoustic  velocity  differs  be- 
tween bone  and  soft  tissue,  variations  in  skull 
thickness  will  produce  relative  changes  in  the 
phases  of  the  acoustic  wavelets  emerging  from  the 
inner  table  of  the  skull.    The  velocity  of  sound 
has  been  reported  to  be  approximately  3050  m/s 
in  skull  bone  (Martin  and  McElhaney  [9])  while 
the  velocity  in  brain  has  been  measured  to  be  ap- 
proximately 1540  m/s  (Wells  [10]).    The  phase 
variation  H  can  be  described  by 

^^-_2^f(pL-m  (1) 

where  Ay  is  the  thickness  variations  of  the  skull 
bone;  Cs  is  the  velocity  of  sound  in  the  skull 


210 


1 


TRANSDUCER  SKULL 


AY 

Fig.  2.    Phase  aberrations  introduced  by  the 
presence  of  skul  1 . 

bone  and  Cg  is  the  velocity  of  sound  in  brain. 

Based  on  the  attenuation  data  and  skull  thick- 
ness variation,  the  skull  bone  can  be  modeled  as 
an  attenuating  plate  of  varying  thickness  posi- 
tioned in  front  of  the  transducer  array.  The 
validity  of  this  model  has  been  demonstrated  by 
studies  of  acoustic  transmission  through  skull 
flaps  performed  in  a  water  tank  environment  with 
a  linear  array  transducer  (Phillips  et  al .  [8]). 
Acoustic  field  plots,  experimentally  measured 
through  skull,  were  compared  with  calculated 
field  plots  using  thickness  data  from  the  same 
skulls.    If  the  proposed  model  of  the  skull  is 
valid  as  a  first  order  approximation,  then  the 
character  of  the  calculated  plots  should  corre- 
late closely  with  the  experimental  field  measure- 
ments.   Sections  of  skull  from  the  temporal  re- 
gion were  frozen  upon  removal  ,  thawed  before 
experimental  use  and  then  continuously  stored  in 
water.    A  diagram  of  the  experimental  arrangement 
is  shown  in  figure  3.    The  multielement  array 
measured  24  mm  in  length  and  consists  of  16, 
equally-spaced,  transducer  elements.    Each  ele- 
ment is  0.35  mm  in  width  and  14  mm  in  elevation. 
The  center  frequency  was  experimentally  measured 
to  be  1.8  MHz  in  water.    The  linear  array  was 
placed  just  under  the  surface  of  the  water  with 
the  skull  flap  positioned  in  close  proximity. 
Independent  control  of  the  transmit  timing  delay 
for  each  element  was  provided  by  continuously 
variable,  mono-stable  multivibrators  whose  out- 
puts were  directly  coupled  into  the  solid  state 
transmitter  pulsers.    The  acoustic  field  patterns 
of  the  array  were  recorded  using  a  small  trans- 
ducer probe  which  could  resolve  small  spatial 
variations  of  acoustic  pressure.    A  Helix  trans- 
ducer probe  measuring  0.46  mm  in  diameter  was 
used  for  this  purpose.    The  acoustic  pressure 
was  displayed  on  a  cathode  ray  tube  and  recorded 
on  photographic  film  as  a  function  of  the  later- 
al position  of  the  translated  field  probe. 

The  Helix  probe  was  then  moved  aside  and  a 
simple  piston  transducer  12.5  mm  in  diameter  was 
used  as  an  illuminating  transducer  which  provided 


TRANSDUCER 


Fig.  3.    Measurement  system  for  transducer  field 
plots . 

a  smooth  acoustic  wavefront  incident  at  the  inner 
table  of  the  skull.    The  relative  arrival  times 
of  the  acoustic  energy  were  recorded  at  the  in- 
dividual array  elements.    The  arrival  time  or 
phase  variations  are  directly  related  to  the 
variations  in  thickness  of  the  skull  across  the 
transducer  aperture.    As  a  theoretical  check  of 
the  proposed  model  of  skull  bone,  computer  simu- 
lations of  acoustic  field  pressure  were  con- 
structed using  the  relative  phase  shifts  of  the 
acoustic  pulses  which  had  traveled  through  the 
skull  to  the  individual  array  elements. 

Figure  4  compares  the  experimental  and  cal- 
culated acoustic  field  plots  which  are  used  to 
evaluate  the  ultrasound  beam  character.  Relative 
acoustic  pressure  was  normalized  to  the  spatial 
maximum  of  each  scan  and  plotted  as  a  function 
of  lateral  position.    The  extent  of  the  lateral 
translation  is  25  mm.    The  element  transmit  de- 
lays were  set  for  a  10  cm  Gaussian  focus,  and 
the  experimental  field  plot  was  recorded  in  the 
focal  plane. 

The  upper  plot  shows  the  measured  acoustic 
pressure  at  a  range  of  10  cm  without  the  presence 
of  a  skull  flap;  the  middle  plot  shows  the  meas- 
ured acoustic  pressure  when  the  adult  skull  is 
placed  in  front  of  the  linear  array.    Of  parti- 
cular interest  is  an  azimuthal  shift  of  the  main 
lobe  to  the  right  and  the  broad  main  lobe  beam- 
width.    The  bottom  plot  shows  the  computer  cal- 
culation where  the  experimentally  measured  ar- 
rival times  were  incorporated  into  the  transmit 
timing  sequence  to  simulate  the  acoustic  field 
pattern.    In  each  simulation  a  transmit  pulse 
character  was  specified  along  with  an  attenua- 
tion factor  for  each  element  to  more  accurately 
represent  the  experimental  situation.  However, 
the  amplitude  variations  across  the  transducer 
aperture  appeared  to  be  of  little  significance 
for  thickness  variations  on  the  order  of  ultra- 
sonic wavelengths.    If  the  proposed  model  of  the 
skull  is  valid  as  a  first  order  approximation, 
then  the  character  of  the  simulation  should  cor- 
relate closely  with  the  acoustic  field  plot 
obtained  experimentally.    The  computer  simula- 
tion shows  an  azimuthal  shift  of  4.9  mm  to  the 


211 


0 

1 1  — 

10  0 

1  i  8 

 1  1  

f  u 

1 

/I 

1  1 

ft          1  9 

0             1  L 

UJ 

1.0 

THROUGH 

'LITUC 

.75 

.  SKULL 

2: 

.50 

UJ 

> 

1— 
■a: 
_] 

UJ 

.25 

0 

1  1 

1  I 

1 

1  1 

12  8 

4  0 

4 

8  12 

PHASE 

COMPENSATION 


Fig.  4.  Comparison  of  experimental  versus  cal- 
culated transducer  field  plots  through 
the  skull. 

right  compared  with  the  5.0  mm  shift  in  the 
experimental  plot.    The  6  dB  beamwidths  in  both 
plots  are  noted  to  be  7.0  mm.    The  overall  geo- 
metrical similarities  lend  considerable  support 
to  the  proposed  first  order  model  of  the  skull. 

4.    Resolution  Limitations  in  the 
Presence  of  Phase  Variations 

One  should  now  consider  what  is  the  extent  of 
image  degradation  due  to  skull  thickness  varia- 
tion in  B-mode  echoencephal ography .    As  has  been 
briefly  discussed  in  a  previous  publication 
(Smith  et  al.  [7]),  the  skull  thickness  itself 
and  hence  the  phase  across  a  one-dimensional 


transducer  can  be  written  as  an 
polynomial : 


,th 


order 


A  + 


C2 

x 


D3  +  E"^ 
X  x 


(2) 


where  <^  is  the  phase  variation  and  x  is  the  loca- 
tion on  the  skull  or  transducer.    Thus  the  phase 
variation  due  to  the  skull  can  be  described  as 
the  sum  of  a  mean  phase  shift,  a  linear  phase 
variation  and  higher  order  terms.    On  the  average 
the  magnitude  of  the  coefficients  will  decrease 
with  higher  orders,  i.e.,  the  finer  grain  varia- 


tions will  have  smaller  amplitudes  than,  for  in- 
stance, the  linear  thickness  variation  of  the 
skull  across  the  transducer.    It  is  also  evident 
that  the  thickness  variations  of  the  skull  will 
have  less  significance  for  imaging  systems  using 
longer  wavelengths. 

The  mean  thickness  and  the  1st  order  thickness 
variation  of  the  skull  make  up  the  first  two 
terms  of  the  phase  polynomial.    The  mean  thick- 
ness of  the  skull  with  its  acoustic  velocity  of 
3050  m/s  results  in  a  range  shift  in  an  individ- 
ual A-mode  or  B-mode  line.    A  linear  thickness 
variation  of  the  skull  results  in  a  refraction 
of  an  A-mode  or  B-mode  line  according  to  Snell 's 
law.    A  range  shift  or  an  azimuthal  shift  does 
not  alter  the  beam  width  of  a  transducer.  Hence, 
these  effects  do  not  degrade  the  resolution  of  an 
A-mode  line  or  individual  line  of  a  B-mode  image. 
However,  for  a  linear  B-scan  or  a  compound  B-scan, 
a  linear  skull  phase  variation  which  changes  as 
a  transducer  is  moved  across  the  skull  can  cause 
distortions  and  misregistration  in  both  the 
range  and  lateral  dimensions.    A  phased  array 
sector  scanner  as  described  by  Somer  [11]  or 
Thurstone  and  von  Ramm  [12]  produces  a  B-scan 
image  in  real  time  through  a  single  fixed  spot 
on  the  skull.    Targets  comprising  such  a  sector 
image  are  shifted  uniformly  in  range  and  azimuth 
by  the  first  two  terms  of  the  phase  polynomial 
and  consequently  there  is  no  image  distortion. 

Second  and  higher  order  skull  phase  variations 
across  the  transducer  aperture  act  like  an  unde- 
sired  lens  and  degrade  the  lateral  resolution  of 
a  single  A-mode  line  and  every  type  of  B-mode 
scanner.    The  more  random  the  phase  variations, 
the  more  disrupted  is  the  diffraction  pattern  of 
the  transducer.    In  fact,  an  upper  limit  can  be 
put  on  the  resolution  capability  of  conventional 
imaging  through  a  phase  aberrating  medium  (Good- 
man [13]).    If  A  is  the  average  linear  dimension 
of  the  transducer  over  which  the  phase  aberra- 
tion is  constant  to  within,  say,  one  radian, 
then  the  resolution  capability  of  the  system  is 
approximately  that  of  a  diffraction  limited  sys- 
tem with  an  effective  transducer  aperture,  Dg^^, 
related  to  A  by 


D 


eff 


A 


(3) 


where  F  is  the  distance  from  the  transducer  to 
its  focal  point,  and  p  is  the  distance  from  the 
transducer  to  the  aberrating  medium.  Transducer 
apertures  of  size  greater  than  Deff  collect  more 
energy  but  have  no  greater  resolution  capability. 
Note  that  the  most  severe  resolution  limitation 
occurs  for  p  -  0.  that  is,  when  the  perturbing 
medium  is  directly  adjacent  to  the  transducer 
in  which  case  Dgff  =  A.    Thus,  a  contact  scanner, 
no  matter  what  its  aperture  size,  is  limited  to 
the  same  resolution  as  a  transducer  of  size  A. 
Long  path,  focussed,  mechanical  water  bath  scan- 
ners such  as  have  been  described  by  Fry  et  al . 
[1]  and  Thurstone  et  al .  [14]  can  maintain 
larger  effective  apertures,  since  the  trans- 
ducer is  sufficiently  removed  from  the  head  so 
that  the  surface  area  of  the  skull  subtended  by 
the  beam  is  very  small  and  the  relative  phase 
variation  is  less  significant.    However,  these 
mechanical  scanners  must  move  over  the  surface 
of  the  skull  making  them  susceptible  to  reg- 
istration errors  due  to  refraction  effects 


212 


from  linear  phase  variations  as  explained  above. 
Tomograms  from  a  contact  electronic  sector  scan- 
ner suffer  no  distortions  from  linear  phase 
variations.    Therefore,  if  much  of  the  thick- 
ness variation  of  the  skull  is  a  linear  varia- 
tion, an  electronic  sector  scanner  will  be  able 
to  maintain  much  of  its  normal  resolution 
capabi 1 ity . 

One  should  now  consider  how  large  the  thick- 
ness variation  or  phase  variation  is  in  the 
skull  in  the  areas  most  commonly  used  for 
acoustic  windows. 

For  a  conventional  imaging  system,  the  amount 
of  deviation  of  <fi(x)  from  the  constant  value  A, 
in  eq.  (2) ,  wi 1 1  be  a  good  predictor  of  how  badly 
the  image  resolution  will  be  degraded  by  phase 
aberrations . 

Fried  [15]  has  considered  this  question  for 
astronomical  imaging,  wherein  the  changing  re- 
fractive index  of  the  moving  air  causes  a  phase 
aberration  affecting  optical  telescopes.  One 
function  which  can  be  used  to  predict  resolution 
limitation  is  the  root  mean  square  phase  varia- 
tion, normalized  to  aperture  size  plotted  as  a 
function  of  aperture  size. 


D/2 

/  [ 


<j;(x)  -  ^ury 


-D/2 


dx 


(4) 


<  >  is  an  ensemble  average,  T(xT  is  the  mean  phase 
across  the  aperture,  and  (()(x)  is  the  phase  shift 
due  to  the  skull  at  some  point,  x,  within  the 
aperture.    To  examine  this  function,  phase  varia- 
tions were  measured  for  the  16  elements  of  a 
linear  array  which  are  separated  by  1.5  mm.  Sets 
of  readings  were  taken  at  10  different  positions 
in  the  temporal  regions  from  an  adult  skull  and  a 
pediatric  skull.    The  measurements  were  made  by 
recording  the  time  of  flight  of  an  acoustic  pulse 
as  it  propagates  through  the  skull  and  arrives  at 
the  elements  of  the  linear  array. 

For  a  given  aperture  size,  the  mean  phase  was 
first  determined,  and  then  the  squares  of  the  de- 
viations were  calculated  for  each  element  within 
that  aperture.    The  sum  of  the  squares  was  taken 
and  then  normalized  to  the  aperture  size.  This 
was  done  for  every  possible  aperture  of  elements 
within  one  set  of  measurements  e.g. ,  for  an  aper- 
ture size  of  eight  elements  there  are  nine  pos- 
sible apertures  within  each  16  element  array. 
The  procedure  was  performed  for  each  of  the  ten 
sets  of  measurements;  a  final  ensemble  average 
was  made,  and  the  square  root  was  taken. 

Figure  5  shows  the  result  of  the  calculations 
for  two  frequencies.    The  root  mean  square  phase 
variation  is  plotted  in  terms  of  radians  versus 
aperture  size  for  1.8  MHz  and  1  MHz.    The  func- 
tion is  monotonically  increasing  at  least  up  to 
an  aperture  size  of  24  mm.    It  will  level  off  at 
some  larger  size  since  the  maximum  thickness 
variations  of  the  bone  are  not  more  than  a  couple 
of  millimeters. 

Fried  [16]  chooses  the  criterion  that  for  root 
mean  square  (rms)  phase  variations  of  1  radian 
(a/2tt),  the  resolution  capability  has  reached  an 
upper  limit  no  matter  what  the  aperture  size. 
Increasing  the  aperture  dimension  beyond  this 
point,  increases  the  phase  variation  so  rapidly 
that  many  elements  on  the  aperture  are  always  out 
of  phase.    From  figure  5,  the  phase  variation  is 


8  16  24 

APERTURE  SIZE  (mm) 

Fig.  5.    Root  mean  square  skull  phase  deviation 
from  mean  phase  shift  versus  aperture 
size. 

one  radian  for  an  aperture  of  6  mm  at  1.8  MHz. 
Therefore,  the  maximum  effective  aperture  for  a 
conventional  contact  ultrasonic  scanner  at  1.8 
MHz  is  6  mm,  i.e.,  a  numerical  aperture  of  7.3 
wavelengths  (ratio  of  aperture  size  to  wave- 
length) and  a  resolution  capability  of  7.8  de- 
grees using  the  Rayleigh  criterion.    At  1  MHz 
the  rms  phase  variation  reaches  one  radian  for 
an  aperture  size  of  18  mm.    Therefore,  the  maxi- 
mum effective  aperture  for  a  conventional  con- 
tact ultrasonic  scanner  at  1  MHz  is  18  mm  or  a 
numerical  aperture  of  12.2  wavelengths  and  a  re- 
solution capability  of  4.6  degrees.    One  con- 
cludes then  that  in  the  presence  of  such  skull 
variations  a  conventional  contact  B-scanner  can 
achieve  a  larger  useful  numerical  aperture  and 
hence  better  lateral  resolution  at  1  MHz  than 
at  1.8  MHz. 

At  this  point  it  would  be  interesting  to  make 
another  set  of  calculations  which  would  be  ap- 
plicable to  an  electronic  sector  scanner  using 
a  contact  transducer.    It  has  been  mentioned  that 
such  an  imaging  system  is  insensitive  to  linear 
phase  variations  across  its  aperture.  Conse- 
quently, the  relevant  function  to  be  calculated 
is  the  rms  phase  deviation,  R^,  from  the  phase 
slope  for  a  given  aperture  size;  i.e.,  the  de- 
viation from 


<^l{x)  =  A  +  Bx 


(5) 


Following  the  same  line  of  thought  as  above,  we 
now  calculate 


D/2 


-D/2 


(t)(x)    -  (f>|^(x) 


dx  > , 


(6) 


where  (j)|_(x)  is  the  slope  of  the  phase  data  for  a 
given  aperture  size.    In  this  series  of  calcula- 
tions, using  the  same  data  as  above,  the  slope 
and  intercept  were  determined  by  the  least  squares 
methods,  and  the  calculations  proceeded  as  above. 
Figure  6  shows  the  results  to  be  as  expected. 
For  an  electronic  sector  scanner  at  1.8  MHz,  the 
rms  phase  variation  reaches  a  critical  value  of 


213 


LU 

a. 
o 

Ui  1.5r 


APERTURE  SIZE  (mm) 

Fig.  6.    Root  mean  square  skull  phase  deviation 
from  phase  slope  versus  aperture  size. 

one  radian  at  an  aperture  size  of  approximately 
14  mm.    Therefore,    on  the  average,  the  maximum 
effective  aperture  size  will  be  14  mm  or  17  wave- 
lengths and  the  resolution  capability  will  be 
3.4  degrees. 

Now  at  1  MHz,  the  rms  phase  variation  for  a 
sector  scanner  never  reaches  one  radian  for  the 


Fig.  7.    Horizontal  cross  sectional  ultrasound 

image  of  brain  through  adult  skull  com- 
pared to  anatomical  brain  section. 


aperture  data  which  we  measured.    If  we  extrap- 
olate this  curve,  we  find  that  it  will  reach  one 
radian  near  1.25  inches  or  32  mm.    This  cor- 
responds to  a  maximum  effective  numerical  aper- 
ture of  21  wavelengths  and  a  resolution  capa- 
bility of  2.7  degrees. 

These  calculations,  based  on  limited  data, 
have  yielded  consistent  results.    For  the  one- 
dimensional  skull  data  analyzed  here,  a  1  MHz 
transducer  can  achieve  a  larger  effective  numer- 
ical aperture  and  hence  better  lateral  resolu- 
tion than  can  a  transducer  operating  at  a  higher 
frequency.    The  considerations  of  the  increasing 
attenuation  with  frequency  of  ultrasound  by  the 
skull  lend  added  support  to  the  choice  of  a 
1  MHz  transducer  for  transkull  imaging.    In  ad- 
dition since  a  phased  array  sector  scanner  is 
insensitive  to  linear  phase  variations  due  to 
the  skull,  such  a  system  would  offer  advantages 
in  maintaining  its  resolution  capability  over 
conventional  B-scanners. 

A  representative  image  is  shown  in  figure  7 
using  the  real  time  phased  array  B-scan  system 
with  a  1  MHz,  31  mm  linear  array  transducer. 
A  horizontal  ultrasound  tomogram  through  the 
skull  of  a  45  year  old  female  is  compared  to  an 
anatomical  cross-section  (Roberts  and  Hanaway 
[17]).    As  indicated,  one  can  see  the  far  skull, 
the  mid-cerebral  fissure,  the  posterior  areas 
of  the  lateral  ventricles,  mid-line  structures 
and  echoes  from  the  sylvian  fissure.    The  corpus 
callosum  appears  as  the  relatively  echo  free 
band  between  the  mid-cerebral  fissure  and  the 
posterior  areas  of  the  lateral  ventricles. 
Images  of  similar  quality  along  with  real  time 


anterior  horns  of 
lateral  ventricles 


Fig.  8.    Coronal  cross  sectional  ultrasound  image 
of  brain  through  skull  compared  to  ana- 
tomical brain  section. 


214 


display  or  pulsating  cephalic  blood  vessels  have 
been  obtained  in  both  horizontal  and  coronal 
sections . 

Figure  8  shows  a  coronal  scan  in  a  ten  year 
old  normal  female  taken  slightly  posterior  to 
the  pinna  of  the  ear.    In  this  view  the  far 
skull,  the  mid-cerebral  fissure,  the  anterior 
horns  of  the  lateral  ventricles,  the  third 
ventricle  and  the  near  side  sylvian  fissure  are 
visualized.    Furthermore,  the  corpus  callosum 
is  again  seen  as  a  relatively  echo-free  band 
above  the  anterior  portions  of  the  lateral 
ventricles.    In  this  view  the  posterior  cerebral 
arteries  are  consistently  seen  as  pulsating 
echoes  in  the  lower  portion  of  the  scan  on  both 
sides  of  the  mid-line.    In  addition,  the  branch 
of  the  middle  cerebral  artery  is  also  seen  quite 
frequently  in  the  area  of  the  sylvian  fissure. 

5.    Resolution  Improvement  via 
Phase  Compensation 

Having  arrived  at  an  improved  transducer  con- 
figuration to  overcome  partially  the  attenuation 
and  phase  aberration  effects  of  the  skull,  signal 
processing  techniques  have  been  used  in  prelimi- 
nary experiments  to  remove  more  completely  the 
effects  of  skull  thickness  variations  in  a  phased 
array  imaging  system. 

Perhaps  the  most  logical  signal  processing 
method  is  to  measure  the  phase  variation  <i>{x)  of 
the  aberrating-medium  before  an  image  is  made, 
determine  a  compensating  phase  variation,  -<t>(x), 
and  then  add  such  a  phase  compensation  to  the 
imaging  system  in  the  form  of  an  acoustic  lens. 
Independent  control  over  each  element  of  the 
transducer  array  in  transmit  and  receive  modes 
provided  a  method  to  produce  any  desired  wave- 
front  response  within  the  limitations  imposed  by 
element  size  and  spacing  with  respect  to  the 
wavelength  of  ultrasound  used.    Such  a  process 
has  long  been  recommended  to  compensate  for  lens 
aberrations  in  optical  telescopes  (Tsujiuchi 
[18]). 

Recently  several  groups  have  reported  con- 
struction of  large  sampled  aperture  telescopes 
whose  mirror  elements  will  move  to  compensate 
for  the  time  varying  phase  aberrations  of  the 
turbulent  atmosphere  in  real  time  (Muller  and 
Buffington  [19];  Hardy  et  a1 .  [20]). 

Figure  9  illustrates  the  basic  principle  of 
the  phase  compensation  technique  in  an  elec- 
tronic phased  array  scanner  (Phillips  et  al . 
[8]).    Five  representative  elements  of  the  trans- 
ducer  array  are  chosen  with  a  section  of  skull 
placed  in  front  of  them.    Part  A  shows  the  phase 
aberrated  wavefronts  emerging  from  a  variable 
thickness  skull  when  the  elements  are  phased  to 
produce  a  focussed  wavefront  in  a  medium  of 
constant  acoustic  velocity.    In  B,  an  acoustic 
wavefront  provided  by  another  source  passes  from 
the  inner  table  of  the  skull  to  the  transducer 
elements.    Knowledge  of  the  spatial  character 
of  the  incident  wavefront  and  the  arrival  times 
at  each  transducer  element  allows  for  the  de- 
termination of  the  relative  changes  in  skull 
thickness  in  front  of  the  elements  comprising 
the  aperture.    When  these  relative  phase  varia- 
tions are  incorporated  into  the  transmit  timing 
as  shown  in  C,  the  emergent  wavefronts  exhibit 
a  restored  phase  character  similar  to  that 
originally  intended. 


received  echo  programmable 


Fig.  9.    Principle  of  the  phase  compensation 
technique. 

The  following  experiment  using  the  equipment 
configuration  of  figure  3  is  presented  to  il- 
lustrate preliminary  findings  (Phillips  et  al . 
[8])  utilizing  phase  compensation.    As  shown  in 
figure  10a  the  control  field  plot  was  recorded 
at  a  range  of  10  cm  when  the  transducer  elements 
were  phased  for  a  10  cm  focus  and  the  skull  re- 
moved.   The  6  dB  beamwidth  was  experimentally 
measured  to  be  4.5  mm  compared  with  a  calculated 
4.3  beamwidth  for  the  diffraction  limited  aper- 
ture.   Figure  10b  shows  the  beam  plot  with  the 
adult  skull  positioned  between  the  array  and 
field  probe.    The  6  dB  beamwidth  was  increased 
to  14.0  mm,  and  a  main  lobe  refraction  of  3.0  mm 
to  the  right  is  also  apparent.    The  relative 
phase  changes  as  a  function  of  skull  position 
in  front  of  the  linear  array  were  measured,  and 
the  phase  compensated  beam  profile  is  shown  in 
figure  10c.    Although  the  acoustic  beam  is  at- 
tenuated due  to  a  single  transmission  through 
skull  bone,  the  phase  character  due  to  variations 
in  skull  thickness  is  restored.    The  refracted 
lobe  was  returned  to  within  0.5  mm  of  the  axis 
defined  in  the  control,  and  the  6  dB  beamwidth 
was  significantly  improved  and  measured  to  be 
4.5  cm.    Although  the  foregoing  studies  were 
performed  for  the  transmit  mode  in  a  direction 
normal  to  the  face  of  the  transducer,  it  was 
hoped  that  similar  improvements  could  be  real- 
ized when  the  ultrasound  pulse  was  electroni- 
cally steered  and  focussed  in  the  receive  mode. 

The  next  step  was  to  perform  phase  com- 
pensation studies  for  the  transmit-receive 
mode  in  the  water  tank  environment  through  a 


215 


mm 


Fig.  10.  Acoustic  field  plots  demonstrating  phase 
compensation  in  the  transmit  mode. 

preserved  skull  segment.    A  phased  array  B-scan 
system  capable  of  2  to  4  mm  resolution  in  range 
and  azimuth  throughout  a  15  cm  field  of  view 
was  used  (von  Ramm  and  Thurstone  [3]).    A  31  mm, 
1  MHz,  linear  array  was  used  to  minimize  the  ef- 
fects of  skull  attenuation  and  reduce  the  rela- 
tive phase  variation  across  the  transducer  as  has 
been  described  above.    To  implement  the  phase 
compensation  technique,  pulses  from  the  array 
passed  through  the  skull  and  were  reflected  by  a 
wire  target  located  at  a  range  of  7  cm.  Not- 
withstanding the  aberrated  character  of  the  ini- 
tial transmitted  pulse,  the  echoes  from  the  wire 
target  return  to  the  skull  as  spherical  waves 
since  the  wire  approximates  a  line  reflector  of 
infinitesimal  size. 

Figure  11  illustrates  the  process  of  phase 
compensation.    Part  A  in  figure  11  shows  the  re- 
ceived echoes  for  five  representative  channels. 
The  returning  echoes  arrive  at  individual  ele- 
ments of  the  array  according  to  the  spherical 
character  of  the  incident  wavefront.    Since  the 
programmable  delay  lines  are  sequenced  to  pro- 
vide a  focus  for  all  points  throughout  the  ob- 
ject volume  of  interest,  the  electronic  pulses 
corresponding  to  the  received  echoes  emerge  from 
the  delay  lines  at  identical  times  as  shown  to 
the  right.    In  part  B  a  skull  sample  was  placed 
in  front  of  the  linear  array.    Due  to  the  aber- 
rating nature  of  the  skull  the  emerging  wavefront 
no  longer  resembles  a  diverging  spherical  wave. 
Since  the  delay  times  are  sequenced  for  a  focus 
in  a  homogeneous  medium  it  is  not  surprising  that 


Fig.  11.  Principle  of  phase  compensation  tech- 
nique in  the  receive  mode. 

the  outputs  from  the  delay  lines  are  unable  to 
provide  phase  coherence  for  an  echo  returning 
from  a  point  target  when  the  skull  is  interposed. 
In  part  C  the  received  echoes  from  the  same  point 
target  through  the  skull  flap  are  restored  to 
phase  coherence  by  adjusting  the  delays  of  each 
delay  line  under  software  control,  so  that  all 
outputs  arrive  in  time  coincidence  with  that  of 
one  channel  chosen  as  reference.    This  procedure 
removes  the  timing  errors  caused  when  imaging 
through  skull  of  varying  thickness  or  composition. 

Results  are  shown  in  the  images  of  figure  12. 
Each  image  was  made  with  the  gain  and  reject 
controls  set  for  optimum  resolution.    The  left 
hand  control  image  shows  the  ultrasound  image  of 
a  line  target  used  to  produce  the  illuminating 
wavefront  and  a  series  of  strings  of  variable 
spacing  (5  mm  to  10  mm)  without  the  presence  of 
a  skull.    The  center  image  shows  clearly  the  de- 
graded resolution  in  the  image  of  the  same  targets 
through  a  section  of  bone  from  the  temporal  area 
of  the  skull.    The  right  hand  image  shows  a  sig- 
nificant restoration  of  image  resolution  for  the 
line  target  and  the  strings  in  the  center  of  the 
field  of  view.    Decreased  off -axis  sensitivity 


216 


CONTROL 


THROUGH  SKULL 


THROUGH  SKULL  WITH 
PHASE  COMPENSATION 


Fig.  12.    Resolution  targets  through  skull  using  phase  compensation. 


through  the  skull  as  the  critical  angle  was  ap- 
proached was  responsible  for  the  target  dropout 
at  the  edges  of  the  field  of  view.  The  phase 
compensation  technique  clearly  reduced  the  spot 
size  of  the  targets  imaged  through  the  skull, 
thus  restoring  much  of  the  azimuthal  resolution 
of  the  system. 

Implementation  of  the  phase  compensation  tech- 
nique in  a  clinical  environment  will  be  more  com- 
plex since  additional  procedures  and  techniques 
must  be  utilized  for  ease  of  clinical  use.  Of 
basic  importance  is  the  necessity  of  producing 
a  smooth  acoustic  wavefront  incident  at  the  inner 
table,  within  a  live,  intact  head.    One  can  specu- 
late that  the  pineal  gland  which  is  calcified  in 
most  adults  or  the  relatively  planar  structures 
of  the  third  ventricle  or  the  far  side  of  the 
skull  may  serve  as  adequate  reflectors  to  use  to 
measure  the  phase  variation  of  the  inner  table 
of  the  skul 1 . 

The  results  presented  here  are  still  prelimi- 
nary.   It  is  difficult  to  predict  how  much  im- 
provement will  be  necessary  in  lateral  resolu- 
tion and  target  dynamic  range  before  B-mode  images 
of  cephalic  structures  routinely  provide  informa- 
tion of  diagnostic  value.    However,  the  improve- 
ments presented  here  are  encouraging  and  may  lead 
to  new  techniques  of  acoustic  imaging  of  cephalic 
structures  having  acceptable  resolution  for  diag- 
nostic evaluation. 

Acknowledgments 

This  work  was  supported  by  USPHS  grants 
HL-12715,  HL-14228,  HS-01613,  and  by  the  Food 
and  Drug  Administration,  Bureau  of  Radiological 
Health. 

References 

[1]      Fry,  F.  J.,  Eggleton,  R.  C. ,  and  Heim- 
burger,  R.  F.,  Transkull  Visualization 
of  Brain  Using  Ultrasound:    An  Experi- 
mental Study,  in  Ultrasonics  in  Medicine, 
M.  de  Vlieger,  D.  N.  White,  and  V.  R. 
McCready,  eds.,  pp.  97-103  (American 
Elsevier  Publishing  Company,  Inc.,  New 
York,  1974). 

[2]      Brinker,  R.  A.  and  Taveras,  J.  M.,  Ultra- 
sound cross-sectional  pictures  of  the  head, 
Acta.  Radiol.  Diag.  5,  745-753  (1966). 


[3]     Von  Ramm,  0.  T.  and  Thurstone,  F.  L., 

Cardiac  imaging  using  a  phased  array  ultra- 
sound system  I:  system  design.  Circulation 
53  (2),  258-262  (1976). 

[ 4 ]     Wh i te ,  D .  N . ,  Ultrasonic  Echo-encephal o- 
graphy .    Medical  Ultrasonic  Laboratory, 
Kingston,  Ontario,  Queen's  University,  1970. 

[5]     Hueter,  T.  F.,  Messung  der  Ul traschal 1  ab- 
sorption in  menschlichen  Schadel knochen  und 
ihre  Abhangigkeit  von  der  Frequenz,  Natur- 
wissenschaften  39,  21-22  (1952). 

[6]     Fry,  F.  J.,  Barger,  J.  E.,  and  Sanghivi, 
N.  T.,  Acoustic  Properties  of  Human  Skull 
Bone.    (Proceedings  of  the  World  Federation 
of  Ultrasound  in  Medicine  and  Biology),  in 
Ultrasound  in  Medicine,  D.  N.  White  and  R. 
Brown,  eds..  Vol.  3,  p.  2095  (Plenum  Press, 
New  York,  1977). 

[7]      Smith,  S.  W.  ,  Phillips,  D.  J.,  von  Ramm, 
0.  T.,  and  Thurstone,  F.  L.,  Real  Time  B- 
mode  Echoencephalography ,  in  Ultrasound  in 
Medicine,  D.  N.  White,  ed..  Vol.  II,  pp. 
373-382  (Plenum  Press,  New  York,  1976). 

[8]      Phillips,  D.  J.,  Smith,  S.  W.,  von  Ramm, 

0.  T.,  and  Thurstone,  F.  L. ,  Sampled  Aper- 
ture Techniques  Applied  to  B-mode  Echo- 
encephalography, in  Acoustical  Holography, 
N.  Booth,  ed..  Vol.  6,  pp.  106-119  (Plenum 
Press,  New  York,  1975). 

[9]      Martin,  B.  and  McElhaney,  J.  N. ,  The 

Acoustic  properties  of  human  skull  bone, 
J.  Biomed  Mater.  Res.  5,  325-333  (1971). 

[10]    Wells,  P.  N.  T.,  Physical  Principles  of 
Ultrasonic  Diagnosis  (Academic  Press,  New 
York,  1969). 

[11]  Somer,  J.  C,  Electronic  sector  scanning 
for  ultrasonic  diagnosis.  Ultrasonics  6, 
153-159  (1968). 

[12]    Thurstone,  F.  L.  and  von  Ramm,  0.  T.  A 

New  Ultrasound  Imaging  Technique  Employing 
Two-Dimensional  Electronic  Beam  Steering, 
in  Acoustical  Holography,  P.  S.  Green,  ed.. 
Vol.  5,  pp.  249-250  (Plenum  Press,  New  York, 
1974). 


217 


[13]    Goodman,  J.  W. ,  Huntley,  W.  H.  Jr.,  Jackson, 
D.  W.,  and  Lehman,  M.,  Wavefront-Reconstruc- 
tion  Imaging  Through  Random  Media,  Appl ied 
Physics  Lett.  8,  311-313  (1966). 

[14]    Thurstone,  F.  L.,  Kjosnes,  N.  I.,  and  Mc- 
Kinney,  W.  M.,  Ultrasonic  scanning  of  bio- 
logic tissue  by  a  new  technique,  Science 
149,-302-303  (1965). 

[15]    Fried,  D.  L.,  Optical  resolution  through  a 
randomly  i nhomogeneous  medium  for  very  long 
and  very  short  exposures,  J.  Opt.  Soc. 
Amer.  56,  1372-1379  (1966). 

[16]    Fried,  D.  L.,  Statistics  of  a  geometric 
representation  of  a  wavefront  distortion, 
J.  Opt.  Soc.  Amer.  55,  1427-1435  (1965). 

[17]    Roberts,  M.  and  Hanaway,  J.,  Atlas  of  the 
Human  Brain  in  Section,  pp.  22-47  (Lea  and 
Feiberger,  Philadelphia,  1970). 

[18]    Tsujiuchi,  J.,  Correction  of  Optical  Images 
of  Compensation  of  Aberrations  and  bv  Spa- 
tial Frequency  Filtering,  in  Progress  in 
Optics,  II,  E.  Wolf,  ed.,  pp.  133-182 
(John  Wiley  and  Sons,  New  York,  1963). 

[19]    Muller,  R.  A.  and  Buffington,  A.,  Real  time 
correction  of  atmospherically  degraded 
telescope  images  through  image  sharpening, 
J.  Opt.  Soc.  Amer.  64,  1200-1210  (1974). 

[20]    Hardy,  J.  W.,  Feibleib,  J.,  and  Wyant, 
J.  C,  Real  Time  Phase  Correction  of 
Optical  Imaging  Systems,  in  Digest  of 
Optical  Propagation  Through  Turbulence 
(abstract  only).    July  9-11,  1974,  Boul der , 
Colorado,  ThBl-l-ThBl-4. 


218 


Chapter  8 
IMAGE  RECONSTRUCTION 


J 


219 


Reprinted  from  Ultrasonic!  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


CHARACTERIZATION  OF  IN  VIVO  BREAST  TISSUE  BY  ULTRASONIC 
TIME-OF-FLIGHT  COMPUTED  TOMOGRAPHY 


G.  H.  Glover 

General  Electric  Company 
Medical  Systems  Division 
Applied  Science  and  Diagnostic  Imaging  Lab 
Milwaukee,  Wisconsin    53201,  U.S.A. 


The  use  of  ultrasonic  time-of-f 1 ight  (TOF)  computed  tomography  for  characteriza- 
tion of  tissue  in  live  breasts  is  reported.    Quantitative  distributions  of  the  re- 
fractive index  within  a  tomogram  of  the  specimen  were  obtained  by  reconstruction  from 
5  MHz  TOF  projection  data.    Fifteen  breast  cancer  patients  and  five  asymptomatic 
volunteers  were  scanned  during  the  clinical  feasibility  study.    The  results  indicate 
that  young,  dense  breasts  have  wide  variation  in  the  refractive  index  distributions. 
In  older  subjects,  however,  various  lesions  are  found  to  have  distinctive  indices. 
Histograms  of  the  reconstructions  show  differences  between  pathological  and  normal 
breasts. 

Keywords:    Breast  cancer;  computerized  tomography;  mammography;  time-of-f 1 ight; 
tissue  characterization;  ultrasonic  imaging;  ultrasound. 


1.  Introduction 

Ultrasonic  energy  potentially  provides  an  at- 
tractive modality  for  mass  screening  programs 
in  breast  cancer  detection.    Its  advantage  is 
particularly  evident  in  light  of  the  recent 
publicity  concerning  suspected  hazards  of  radio- 
graphic procedures.    It  is,  therefore,  of  in- 
terest to  characterize  quantitatively  the  ultra- 
sonic properties  of  breast  tissue,  with  an  eye 
towards  differentiation  of  various  neoplasms  and 
breast  parenchyma.    A  useful  vehicle  for  this 
purpose  (and  perhaps  for  mass  screening  as  well) 
is  computerized  time-of-f 1 i ght  (TOF)  tomography. 

TOF  tomography  was  first  published  by  Green- 
leaf  et  al.  [l]i  in  1975.    Since  then,  several 
other  workers  [2,3]  have  published  reports  on 
other  aspects  of  this  technology.    In  this  tech- 
nique, the  speed  of  propagation  of  an  ultrasonic 
wave  is  accurately  measured  along  many  uniformly 
sampled  paths  through  the  specimen  (see  fig.  1). 
The  data  are  obtained  from  measurements  of  the 
time  delay  for  reception  of  a  short  pulse  travers- 
ing the  path.    In  the  parallel-scan  geometry 
shown  in  figure  1,  the  set  of  values  corresponding 
to  a  given  angle  relative  to  the  specimen  collec- 
tively form  one  TOF  projection  (or  view).  Other 
projections  are  then  obtained  by  rotating  the 
scanner  relative  to  the  specimen.    A  two- 
dimensional  refractive  index  distribution  is  then 
obtained  from  the  projection  data  by  computerized 
image  reconstruction  techniques  [4].    The  result- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


SIDE  VIEW 


Fig.  1.    Geometry  of  scanner.    A  single  pair  of 
transducers  are  translated  in  1  mm  steps 
to  form  one  projection.    One  hundred 
projections  are  obtained  by  rotation  of 
the  scanner  through  180  degrees.  Other 
slices  are  acquired  with  the  scanner 
lowered  relative  to  the  breast  (side  view). 


221 


ing  images  are  displayed  as  an  array  of  square 
pixels,  each  of  which  represents  the  local  speed 
of  propagation  supported  by  the  tissue.  The 
process  is  very  much  analogous  to  x-ray  computed 
tomography  methods. 

The  speed  of  propagation  of  an  ultrasound  wave 
depends  upon  the  elasticity  and  mass  density  of 
the  propagating  medium.    Thus,  the  TOF  reconstruc- 
tions directly  provide  quantitative,  localized  in- 
formation about  significant  mechanical  properties 
of  the  bulk  tissue.     (By  contrast,  echo  amplitude 
is  a  very  complicated  function  of  interfacial  and 
bulk  tissue  ultrasonic  characteristics,  as  well 
as  of  geometry.)    The  propagation  velocity  has 
been  found  to  vary  by  several  percent  in  various 
types  of  soft  tissue  [5].    In  this  paper,  velocity 
measurements  of  tissue  structure  in  in  vivo  breasts 
are  presented  and  analyzed  with  the  aid  of  histo- 
grams.   In  the  following  section,  the  experimental 
technique  is  described. 

2.  Experiment 

A.  Technique 

Only  a  brief  description  is  given  here  as  the 
details  have  been  presented  elsewhere  [3,6].  The 
scanner  used  in  the  study  is  shown  in  figure  2.  A 
transmitter  and  a  receiver  transducer  (Panametrics , 
5  MHz)  were  rigidly  mounted  on  a  carriage  which 
could  be  translated  by  a  stepping  motor.  Samples 
were  acquired  at  1  mm  intervals  (163  samples/ 
projection).    The  tank  was  mounted  under  a  canvas 
sling  supported  by  a  metal  and  wooden  structure  so 
that  the  subject  lay  prone  with  the  breast  im- 
mersed in  the  water  of  the  inner  tank.    Thus,  hori- 
zontal slices  5  mm  thick  transaxial  to  the  breast 
were  obtained  as  shown  in  figure  1.    After  the 
first  projection  was  acquired,  additional  projec- 
tions were  measured  as  the  tank  was  rotated  through 
180  degrees  in  1.8  degree  steps  by  a  second  motor. 
The  TOF  data  (n  seconds  of  delay)  were  stored  on 
digital  cassette  tape  for  later  reconstruction. 


Fig.  2.    Scanner  mechanism  mounted  in  wooden  frame. 
Both  tanks  are  water  filled.    The  breast 
is  immersed  in  inner  tank,  which  is  isolat- 
ed for  electrical  precaution. 


The  reconstructions  were  obtained  using  the  con- 
volution algorithm  [4].    The  original  reconstruc- 
tion utilized  time-sharing  software  and  were  dis- 
played in  a  57  x  57  matrix  format  [61.    For  the 
present  work,  the  images  were  re-reconstructed  in 
a  64  x  64  matrix  representing  12  x  12  cm  with  a 
minicomputer.    Reconstruction  time  was  about  60 
seconds  using  conventional  FORTRAN  coding. 

The  velocity  values  v  are  given  in  CTU  values, 
defined  as  [6] 


CTU 


VO 


'0 


x  1000, 


(1) 


where  vg  is  the  velocity  of  propagation  in  the 
water  (at  about  34  °C).    Thus,  +10  CTU  corresponds 
to  a  velocity  1  percent  greater  than  for  water. 

The  clinical  study  was  performed  at  the  Albany 
Medical  College.    Fifteen  symptomatic  patients  and 
five  asymptomatic  volunteers  were  scanned.  Mam- 
mograms and  clinical  reports  for  the  patients  were 
available  before  the  scans  to  help  localize  the 
region  of  interest.    Generally,  three  slices  1  cm 
apart  for  the  pathological  breast  and  one  slice  of 
the  contralateral  breast  were  acquired.    After  the 
TOF  scans,  biopsies  were  performed  on  the  suspi- 
cious breasts.    These  results  were  then  made  avail- 
able for  comparison  with  the  reconstructions.  A 
thorough  historical  comparison  would  have  been 
desirable,  but  was  outside  the  scope  of  this  study. 

B.  Results 

Figures  3  and  4  show  typical  reconstructions  of 
breasts  having  lesions.    In  these  pictures,  high 
velocities  are  depicted  as  black.    Figure  3  shows 
,a  right  breast  (age  53)  containing  a  large  (be- 
nign) fibroadenoma.    The  CTU  values  for  the  lesion 
are  20-28,  while  the  mammory  fat  (white  areas)  has 
CTU  values  of  about  -60.    The  left  breast  depicted 
on  figure  4  (age  44)  contains  infiltrating  ductal 
carcinoma  with  CTU  numbers  40-45.    The  mammory  fat 
has  CTU  values  similar  to  those  in  figure  3. 

The  streaking  artifacts  in  these  pictures  de- 
rive from  patient  motion  during  the  long  scan  time 
required  to  obtain  the  data  (about  9  minutes/ 
slice).    As  a  result,  the  pictures  are  much  noisier 
(standard  deviation,        10  CTU  in  the  water  bath) 
than  the  intrinsic  velocity  resolution  of  the 
system  determined  with  a  stationary  phantom  (a 
2  CTU).    Nevertheless,  the  reconstructions  are  of 
sufficient  accuracy  to  characterize  the  large-mass 
lesions  studied  here. 

Table  1  presents  a  summary  of  the  CTU  values  for 
several  types  of  lesions  as  well  as  fat  and  asymp- 

Table  1.    Summary  of  CTU  values  for  several  types 
of  lesions,  fat  and  asymptomatic  breast 


parenchyma . 

Symptomatic  patients  (ages  33  to  85) 

mammary  fat 

normal  parenchyma 

f i broadenoma/f ibrocystic  disease 

(benign ) 

carcinoma 

inflammatory  skin  thickening 
Asymptomatic  subjects  (ages  24  to  29) 
fibrotic  mammary  glands,  ducts 


CTU  numbers 


-50  to  -70 

0  to  15 

12  to  30 

40  to  60 

40  to  65 


8  to  60 


222 


Fig.  3.    Reconstruction  of  breast  with  large 
fibroadenoma  (black  region). 

tomatic  (presumed  normal)  breast  parenchyma.  In 
the  older  patients,  the  CTU  values  were  found  to 
be  distinctive  for  fat,  normal  parenchyma,  fibrotic 
tissue,  and  carcinoma.    Malignant  lesions  and  in- 
flammatory skin  thickening  due  to  the  presence  of 
an  internal  malignant  neoplasm  have  characteristic 
CTU  numbers  greater  than  40,  while  benign  lesions 
(fibrocystic  disease,  fibroadenoma)  have  CTU  values 
less  than  30.    However,  the  CTU  values  for  the 
dense  breasts  of  the  younger  asymptomatic  subjects 
span  the  entire  range.    It  is,  therefore,  impos- 
sible to  differentiate  solely  on  the  basis  of  CTU 
value  neoplastic  tissue  from  normal  fibrotic 
parenchyma  in  such  cases. 

In  several  slices  wherein  the  lesion  did  not 
completely  fill  the  5  mm  axial  slice  thickness,  the 
CTU  values  were  less  than  the  bulk  numbers  in  the 
table.    These  partial  volume  effects  arise  because 
the  component  of  transmitted  signal  which  arrives 
earlier  due  to  traversal  through  the  high  velocity 
lesion  ultimately  becomes  too  low  in  amplitude  to 
be  detected  relative  to  the  signal  components  which 
exclude  the  lesion.    The  transition  is  not  perfect- 
ly sharp  due  to  refraction  and  tissue  dependent 
attenuation  effects.  Excluding  such  cases,  however, 
there  were  no  counter-examples  to  the  CTU  values 
classifications  in  the  table. 

3.    Statistical  Analysis 

A  rudimentary  statistical  analysis  of  the  re- 
constructions was  performed  by  computing  the 
histograms  and  cumulative  distribution  functions. 
For  this  purpose,  a  program  was  developed  which 
searched  the  matrix  for  the  edge  of  the  breast, 
fit  an  arbitrary  ellipse  to  the  boundary  points 
by  regression,  and  calculated  the  histogram  for 
those  elements  within  the  ellipse.    The  class 
interval  was  chosen  as  3  CTU  numbers  in  com- 
puting the  histograms.    The  results  were  normal- 
ized to  remove  variations  in  specimen  size. 
Figure  5  shows  histograms  for  a  young,  asymp- 


Fig.  4.    Reconstruction  showing  infiltrating  duc- 
tal carcinoma  (lower  quadrant)  and  in- 
flammatory carcinoma  central  region. 

tomatic  subject.    Note  that  the  two  normal 
breasts  have  similar  tissue  distributions.  This 
was  not  always  the  case,  although  generally  the 
skew  and  kurtosis  of  the  histograms  were  similar. 
The  distributions  are  peaked  near  CTU  =  0  and 
contain  considerable  fibrotic  tissue  with  high 
CTU  numbers.    The  variations  in  the  histograms 
for  different  young  subjects  were  quite  wide. 
This  is  illustrated  by  plotting  the  (inverse) 
cumulative  probability  function. 


0j  =  E   fi  (2) 


-80  -60  -40  -20  0  20  40  60 

CTU  VALUES 

Fig.  5.    Histograms  of  right  (solid)  and  left 

(dashed)  breasts  of  24  year  old  asympto- 
matic subject.    The  distributions  are 
very  similar  and  centered  near  CTU  =  0. 
Considerable  fibrotic  structure  is  evident 
from  the  peak  position  and  area  under  the 
high  valued  tails. 


223 


where  f-;  is  the  frequency  for  the  i^'^  CTU  value. 
Curves  for  three  of  the  subjects  are  shown  in 
figure  6.    As  may  be  seen,  the  curves  span  a  large 
range  of  CTU  values. 

10° 


CTU  VALUES 

Fig.  6.    Inverse  cumulative  probability  function 
for  three  normal,  young  subjects.  The 
zero  asymptotes  (which  reflect  peak  CTU 
values)  show  wide  variation. 

Figure  7  shows  histograms  for  a  58  year  old 
patient  with  carcinoma  in  the  right  breast  and  a 
normal  left  breast.    Note  that  the  peak  for  the 
normal  histogram  has  shifted  position  to  low  nega- 
tive values  relative  to  figure  5.    This  reflects 
the  atrophy  of  glandular  structure  (high  CTU)  and 


CTU  VALUES 


Fig.  7.    Histograms  of  58  year  old  patient  with 
carcinoma  in  right  breast  (solid)  and 
normal  (dashed)  left  breast.    The  normal 
distribution  is  peaked  near  CTU  =  -40, 
indicating  atrophy  of  the  glandular  tis- 
sue and  replacement  by  fat.    The  kurtosis 
and  skew  of  the  histogram  for  the  patho- 
logical breast  are  markedly  dissimilar  to 
the  normal  one. 


replacement  by  fat  (low  CTU)  in  the  post-meno- 
pausal  breast.    Although  not  enough  data  is  avail- 
able from  this  study,  undoubtedly  the  peak  posi- 
tion is  a  decreasing  function  of  age  [7].  The 
histogram  for  the  pathological  breast  has  marked- 
ly dissimilar  shape  (skew  and  kurtosis)  and  ex- 
tends to  higher  CTU  values.    The  carcinoma  is 
contained  in  the  high  valued  region.    (The  peak 
near  zero  probably  represents  some  water  being  in- 
cluded in  the  histogram  by  the  algorithm.)  Figure 
8  shows  examples  of  cumulative  distribution  plots 


-80       -60      -40       -20        0         20        40  60 

CTU  VALUES 


Fig.  8.    Inverse  cumulation  probability  function 

for  a  representative  sampling  of  the  older 
patients.    The  zero  asymptotic  values  can 
be  roughly  correlated  with  tissue  type. 

-  -  -  =  normal,    =  fibroceptic/ 

adenoma,    =  carcinoma. 


for  older  normal  breasts,  and  breasts  containing 
fibrotic  and  carcinoma  tissue.    In  general,  figures 
6  and  8  tend  to  reflect  the  results  of  all  the 
scans  summarized  in  table  1. 

4.  Discussion 

Human  female  breasts  are  constituted  of  time 
varying  fractions  of  various  components.    It  ap- 
pears plausible  that  each  of  these  components 
(parenchyma,  glands,  fat,  etc. )  have  ultrasonic 
properties  which  are  relatively  similar  between 
different  subjects.    Indeed,  the  wide  variation  in 
average  and  peak  CTU  numbers  observed  for  young 
dense  breasts  probably  represents  an  incomplete 
analysis  of  the  volumetric  fractions  of  glandular 
tissue,  fat,  and  ducts  present  in  any  given  sub- 
ject.   In  any  case,  a  single  number  representing 
velocity  values  averaged  over  the  volume  of  a 
breast  [7]  is  probably  too  simplistic  to  be  rep- 
resentative of  the  breast. 

The  symptomatic  subjects  examined  in  this  study 
were  chosen  to  have  large,  unambiguous  lesions  in 
order  to  acquire  data  on  the  various  tissue  com- 
ponents.   These  data,  however,  represent  a  statis- 
tically insignificant  number  of  cases  on  which  to 


224 


base  any  absolute  conclusions.    In  fact,  a  larger 
data  base  will  probably  show  wider  variation  in 
values  than  was  found  in  this  study.  Nevertheless, 
the  present  results  suggest  a  distinction  between 
normal  and  neoplastic  tissue  which  is  encouraging 
for  mammographic  application. 

There  are  potentially  several  methods  of  utiliz- 
ing the  TOF  data  in  cancer  detection  suggested  by 
the  histograms.    When  available,  histograms  of  the 
contralateral  breast  provide  a  reference  which 
could  form  part  of  a  comparative  basis.  Alterna- 
tively or  additionally,  a  reference  base  which  de- 
pends on  age  and  other  variables  can  be  accumulated. 
Besides  the  histograms,  which  present  a  volumetric 
picture  of  the  constitutive  elements,  the  individ- 
ual pixel  values,  rates  of  change,  and  geometry  can 
be  searched  for  characteristic  patterns.  Applica- 
tion of  such  speculation,  however,  must  await  the 
acquisition  of  more  massive  amounts  of  data. 

Acknowledgments 

The  clinical  data  was  acquired  with  the  help  of 
R.  W.  Sponzo,  M.D.,  his  staff  and  colleagues  at 
the  Albany  Medical  College. 

References 

[1]    Greenleaf,  J.  F.  et  al . ,  Algebraic  Reconstruc- 
tion of  Spatial  Distributions  of  Acoustic 
Velocities  in  Tissue  from  their  Time  of  Flight 
Profiles,  in  Acoustic  Holography,  W.  Booth, 
ed..  Vol.  VI,  pp.  71-89  (Pergamon  Press,  New 
York,  1975). 


[2]    Carson,  P.  L.,  Oughton,  T.  V.,  and  Hendee, 
W.  R.,  Ultrasonic  Transaxial  Tomography  by 
Reconstruction,  in  Ultrasound  in  Medicine, 
D.  White  and  R.  Barns,  eds..  Vol.  II,  pp. 
341-350  (Plenum  Press,  New  York,  1976). 

[3]    Glover,  G.  H.  and  Sharp,  J.  C,  Reconstruc- 
tion of  Ultrasound  Propagation  Speed  Distribu- 
tions in  Soft  Tissue:    Time-Of-Fl ight  Tomo- 
graphy, IEEE  Trans.  Sonics  and  Ultrasonics 
SU-24,  229-234  (1977). 

[4]    Gordon,  R.  and  Herman,  G.  T.,  Three  dimen- 
sional reconstruction  from  projections:  a 
review  of  algorithms,  Internat.  Rev.  Cytology 
38,  111-151  (1974).   

[5]    See,  e^. ,  Wells,  P.  N.  T. ,  Review:  absorp- 
tion and  dispersion  of  ultrasound  in  biologi- 
cal tissue.  Ultrasound  in  Med,  and  Biol.  1, 
369-376  (1975T:    ~ 

[6]    Glover,  G.  H.,  Computerized  time-of-fl ight 
ultrasonic  tomography  for  breast  examination. 
Ultrasound  in  Med,  and  Biol,  (in  press). 

[7]    Kossoff,  G.,  Fry,  E.  K.,  and  Jellins,  J., 
Average  velocity  of  ultrasound  in  the  human 
female  breast,  J.  Acoust.  Soc.  Am.  53,  1730- 
1736  (1973).  ~ 


225 


I    11    ■  I 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


VARIATION  OF  ACOUSTIC  SPEED  WITH  TEMPERATURE  IN  VARIOUS  EXCISED 
HUMAN  TISSUES  STUDIED  BY  ULTRASOUND  COMPUTERIZED  TOMOGRAPHY 


B.  Rajagopalan,  J.  F.  Greenleaf,  P.  J.  Thomas, 
S.  A.  Johnson,  and  R.  C.  Bahn 

Biodynamics  Research  Unit 
Departments  of  Physiology  and  Biophysics,  and 
Department  of  Pathology  and  Anatomy 

Mayo  Clinic  and  Mayo  Foundation 
Rochester,  Minnesota    55901,  U.S.A. 


Variation  of  acoustic  speed  as  a  function  of  temperature  was  measured  in  fresh  ex- 
cised human  liver,  psoas  muscle,  spleen,  spinal  cord,  kidney  and  fat,  parenchyma  and 
muscles  associated  with  female  breasts.    Tissues  were  encased  in  rubber  finger  cots 
and  suspended  in  a  temperature  controlled  water  bath.    A  prototype  clinical  ultrasound 
breast  scanner  was  used  to  obtain  data  required  to  reconstruct  distributions  of  acous- 
tic speed  within  two-dimensional  planes  through  the  tissue  specimens  over  a  tempera- 
ture range  of  20  to  40  °C.    Quantitative  images  (printer  listings  of  acoustic  speed) 
of  64  X  64  pixels  were  used  to  obtain  averages  of  up  to  16  speed  measurements  within 
the  image  of  each  tissue.    The  acoustic  speed  in  most  tissues  increased  monotonical ly 
with  temperature  following  the  behavior  of  water.    Acoustic  speed  of  fat  showed  an 
anomalous  decrease  in  acoustic  speed  around  34  °C  suggesting  possible  phase  transition. 


Key  words:    Aperture  synthesis;  computed  tomography;  Doppler;  fluid  flow;  high 
resolution;  reconstruction;  temperature  reconstruction;  ultrasound. 


1.  Introduction 

Historically,  production  of  images  by  ultra- 
sound scanning  has  been  done  by  displaying  either 
the  pulse  echo  reflected  from  an  object  (B-scan) 
or  by  displaying  the  attenuated  signal  (C-scans) 
transmitted  by  an  object  [l]i.  These  modalities, 
although  useful,  give  only  a  qualitative  mapping 
of  tissue  interfaces  and  geometries.  Images 
representing  quantitative  distributions  of  basic 
mechanical  tissue  properties  as  characterized  by 
ultrasound  scattering  cross  section,  impedance 
[2,3]  and  acoustic  speed  [4]  may  be  intrinsically 
more  valuable  than  the  qualitative  images  rep- 
resenting tissue  borders  and  geometries.  A 
quantitative  imaging  modality  would  eliminate 
operator-dependent  variations  one  often  finds  in 
the  qualitative  imaging  modalities.  Ultrasound 
computerized  tomography  [4]  is  one  such  imaging 
modality  by  which  the  distribution  of  one  funda- 
mental mechanical  parameter,  i.e.,  acoustic 
speed,  in  a  tissue  can  be  studied. 

The  purpose  of  this  report  is  to  describe 
methods  and  initial  results  of  applying  computer- 
ized tomography  to  the  problem  of  obtaining 
quantitative  images  representing  the  distribution 
of  acoustic  speed  and  the  effect  of  temperature 
on  these  values  for  tissues  from  various  parts  of 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


the  body,  especially  the  human  breast.  The 
knowledge  of  temperature  coefficients  of  acoustic 
speed  in  tissues  may  be  useful  in  estimating  the 
temperature  change  in  a  region  of  tissue  occur- 
ring due  to  external  causes  such  as  ultrasonic 
heating  or  drugs.    Since  various  disease  proces- 
ses alter  tissue  temperature,  the  temperature  co- 
efficients may  also  be  helpful  in  the  interpreta- 
tion of  computerized  tomography  images  of  live 
tissues. 

The  reconstruction  problem  in  ultrasound  com- 
puterized tomography  has  been  described  else- 
where [4]  and  will  not  be  described  in  detail 
here.    A  good  introduction  to  the  basic  mathe- 
matics of  computerized  tomography  and  a  general 
overview  of  the  analytical  and  iterative  algo- 
rithms can  be  found  in  the  review  articles  by 
Gordon  and  Herman  [5]  and  Brooks  and  Dichiro  [6]. 
The  convolution  reconstruction  algorithm  for  the 
divergent  beam  geometry  was  first  derived  by 
Lakshminarayanan  [7]  and  later  a  mathematically 
rigorous  treatment  of  the  same  problem  leading  to 
identical  results  was  given  by  Herman  and  his  co- 
workers [8].    This  divergent  beam  convolution  re- 
construction algorithm  is  used  exclusively  for 
all  the  reconstruction  images  in  this  paper. 

2.    Experimental  Arrangement 

In  ultrasound  computerized  tomography  the 
profile  data  consist  of  measurements  of  time-of- 


227 


flight  of  an  acoustic  pulse  through  the  sample 
along  different  rays  or  directions.    The  data 
are  obtained  by  divergent  beam  scanning  with  a 
single  pair  of  transducers  one  on  each  side  of 
the  region  of  interest.    The  schematic  of  the 
data  collection  geometry  is  shown  in  figure  1. 
The  scan  angle  is  a  and  is  incremented  at  0.15° 
intervals  using  a  stepping  motor.    After  the  scan 
the  ultrasound  transmitter-receiver  arm  is  rotat- 
ed to  a  new  e  position  and  another  scan  is  ini- 
tiated.   Thus,  for  each  e  position  there  is  one 
profile  of  time-of-f 1 i ght  data.    One  hundred 
twenty  profiles  separated  in  view  by  3°  increments 
are  gathered  over  the  360°  range. 

transmitter  locus 


— 

profile  data 

V 

P(a,e2) 

Jy\\/ 

tissue]  /  / 

profile  data 
P(a,ei) 


receiver  loci 


Fig.  1.    Data  collection  geometry  for  ultrasound 
tomography,    ej  and  63  are  the  angles 
made  by  the  central  ray  in  the  scan  with 
X-axis  for  two  different  scans.  Profile 
data  P(a,ei)  is  gathered  by  rotating  the 
receiver-transmitter  arm  about  an  axis 
passing  through  the  apex  of  the  sector  and 
perpendicular  to  the  plane  of  the  figure. 
Thus,  the  receiver  loci  are  arcs.  The 
transmitter  mounted  at  the  axis  of  the 
scan  describes  a  circle  as  profile  data 
from  different  views  (e)  are  gathered. 

The  schematic  of  the  experimental  arrangement 
for  collecting  computerized  tomography  data  is 
shown  in  figure  2.    The  system  is  driven  by  an 
Interdata  7/32  minicomputer  which  controls  the 
height  and  rotation  angle  (e)  of  the  scanner  arm 
and  collects  the  time-of-f 1 ight  data  from  the 
digital  output  of  the  time-of-f 1 ight  (TOF)  clock 
and  control  box.    This  TOF  clock  and  control  box 
is  interfaced  to  the  digital  I-O  of  the  mini- 
computer and  upon  a  command  from  the  computer 
collects  time-of-fl ight  data,  outputs  the  data, 
steps  the  scanner  arm  to  the  next  orientation  in 
the  scan  and  repeats  this  process  until  one  profile 
of  data  are  collected.    At  the  end  of  the  scan  the 
scanner  arm  is  rotated  to  the  new  6  position  by  a 
command  from  the  minicomputer  and  the  process  of 
collecting  profile  data  resumed  by  a  trigger  to 
the  TOF  clock  and  control  box.    These  time-of- 
fl  ight  profiles  representing  propagation  delay  of 
acoustic  pulses  along  many  rays  through  the  tissue 
under  examination  are  input  to  the  reconstruction 
programs  to  obtain  quantitative  distributions  of 


keyboard 


height 


control  rotate 


interdata 
7/32 


O 
CRT 


camera 


 ^ 

strobe 


16 


TOF 


signal 


Time-of-Fl ight 
clock  &  control 


receive 


(((V^(( 


^rotatef 

■|hei 

ght| 

trigger| 
transmitf*— ^ 


scanh 


Fig.  2.    Ultrasound  computed  tomography  system: 
Data  collection  and  control.  Computer 
controls  transmi t-del ay-and  position. 
Time-of-f 1 ight  of  signal  is  measured  with 
a  16-bit  100  MHz  clock  and  counter  which 
is  read  by  the  computer  at  up  to  400  points 
along  a  scan  profile,  each  point  separated 
by  0.15°.    One  hundred  views  each  contain- 
ing 400  samples  and  each  sample  containing 
eight  channels  of  16-bit  time-of-f 1 ight 
data  can  be  obtained  in  a  period  of  2.5 
'minutes. 

acoustic  speed  in  the  transverse  plane  through 
the  tissue. 

3.    Reconstruction  Algorithm 

The  mathematical  steps  to  arrive  at  the  recon- 
struction picture  from  the  set  of  profile  data 
are  given  in  this  section  without  attempting  to 
derive  any  of  them.    These  steps  known  as  the  re- 
construction algorithm  are  treated  extensively  in 
the  literature  and  the  derivation  for  divergent 
beam  geometry  starting  from  Radon  transform  [9] 
can  be  found  in  the  reference  mentioned  earlier 
[8]. 

The  raw  data  of  each  profile  consist  of  time 
measurements  (in  seconds  x  10"^)  at  each  incre- 
ment along  the  arc  of  the  scan.    The  data  for  the 
20  or  more  points  closest  to  each  end  of  the  pro- 
file are  the  time- of-fl ight  for  sound  through 
water,  and  never  vary  by  more  than  one  unit 
(10"^  s)  from  the  lowest  to  the  highest.  The 
average  of  the  first  few  points  is  subtracted  from 
each  point  of  the  profile  and  the  differences 
P(j,e)  represent  the  profile  data  for  reconstruc- 
tion.   Here  6  is  the  angle  made  by  the  central  ray 
of  the  scan  with  x-axis  (arbitrarily  chosen)  and 
j  =  -n  . . .  -1 ,  0,  1 ,  . . .  +n  indexes  the  angles  of 
the  2n+l  individual  rays  of  the  scan  with  j  =  0 
at  the  center  of  the  scan.    For  each  rotation 
angle  e  the  profile  data  are  reduced  by  the 
cosines  of  the  angles  of  the  individual  rays  from 
the  center  line. 

P(j,e)  modified  =  P*(j,e)  =  P(j,e)  cos(j£)  (i) 

where  e  is  the  constant  angular  increment  in  the 
scanning  motion  (0.15°).    The  modified  profile 


228 


P*  (j>9)  is  convolved  with  a  kernel  to  give  the 
convolved  modified  profile  C(k,e),  i .e. , 

kmax 

C(k,6)  =  Z  P  (2) 

''"'^min 

The  kernel  is  given  by 

Ao  =  (3) 
A-j  =  0  for  even  i  and  i  =  0 


where  D  is  the  distance  from  the  axis  of  rotation 
for  the  scanning  motion  to  the  axis  of  rotation 
for  rotate  motion  and  kmin  and  kpiax  indicate  the 
lower  and  upper  limits  of  the  scan.  Finally, 
the  back-projection  is  formed,  along  the  lines 
of  the  acoustic  rays,  converging  at  the  axis  of 
scan  motion,  with  the  back-projected  C(k,e) 
weighted  with  inverse  square  of  the  relative 
distance  of  the  point  from  the  axis  of  the  scan 
to  distance  D.    Thus,  for  each  time-of-f 1 ight 
profile  at  an  angle  e  the  matrix  f(x,y)  forming 
the  reconstructed  picture  is  incremented  by 


Af(x,y)e  =  I  C(k,e)  ^  (4) 

where  r  =  distance  of  the  point  (x,y)  from  the 
axis  of  the  scan.    C(k,e)  is  interpolated  if  no 
ray  goes  through  the  point  (x,y).    The  sum  of  all 
the  back-projected  Af(x,y)Q  for  all  e  values 
yields  the  reconstructed  image  at  (x,y). 

f(x,y)=        Z        fC(k,e)Di  (5) 
sum  over  Q 

When  the  back-projection  is  finished,  the  value 
for  each  pixel  will  be  the  increase  or  decrease 
in  time  per  centimeter  at  that  point,  compared 
with  water.    The  time-of-f 1 ight  for  water  (in  the 
same  units,  calculated  from  the  measured  time  and 
known  distance,  and  verified  by  the  known  veloci- 
ty of  sound  in  water  for  the  particular  tempera- 
ture) is  then  added  to  the  value  of  each  pixel, 
and  the  reciprocal  of  the  sum  is  scaled  to  give 
the  final  result  in  meters  per  second.    The  re- 
constructed digital  images  represented  by  a  matrix 
of  64  X  64  pixels  can  be  displayed  in  gray  scale 
in  a  Ramtek  display  system. 

4.    Experimental  Results  and  Discussion 

The  accuracy  with  which  the  acoustic  speed  can 
be  determined  using  ultrasound  computerized  tomo- 
graphy is  demonstrated  in  figure  3  in  the  high 
correlation  of  our  data  with  the  literature  value 
'(10)  for  2.5  percent  salt  solution.    Thin  rubber 
finger  cots  filled  with  saline  served  as  the  test 
object.    The  agreement  of  the  literature  values 
with  the  experimental  results  over  the  entire 
temperature  range  is  within  ±  0.3  percent.  Since 
the  reconstructed  image  is  a  matrix  giving  the 
acoustic  speed  in  each  pixel  (pixel  size  ^  1.5 
mm^)  mean  and  root  mean  square  deviation  of  the 


Fig.  3.    Comparison  of  acoustic  speed  in  2.5  per- 
cent salt  solution  (2.5  g  salt  in  100  cm^ 
of  water),  as  determined  by  ultrasound 
computerized  tomography,  with  the  litera- 
ture values.    The  literature  values  have 
an  accuracy  of  ±  4  cm/s.    Acoustic  speed 
values  were  taken  as  the  average  of  20 
samples  in  the  reconstructed  image. 
oUltrasound  computerized  tomograph  data 
•Millero,  F.  J.  and  Kubinski,  T., 
J_.  Acoust.  Soc.  Am.  57^,  312  (1975). 

acoustic  speed  are  easily  determined  by  analyzing 
the  quantitative  distribution  of  acoustic  speed 
in  the  image.    Similar  statistical  analysis  was 
done  on  the  reconstruction  images  of  all  the  tis- 
sues that  were  studied  and  reported  in  this  paper. 

Tissues  were  obtained  and  selected  by  a  pathol- 
ogist as  being  representative  of  selected  tissue 
types.    Samples  of  tissue  acquired  were  about  2 
to  3  centimeter  square  and  5  centimeter  long  and 
were  suspended  in  a  bath  of  normal  saline.  Ref- 
erence objects  such  as  finger  cots  filled  with 
saline  of  known  concentrations  were  also  suspend- 
ed with  the  tissues.    Later  in  the  study  the 
tissues  were  packed  into  finger  cots  under  normal 
saline  and  suspended  in  the  water  bath.  This 
procedure  simplified  the  tissue  handling  con- 
siderably and  the  effect  of  finger  cots  on  the 
quantitative  results  was  found  to  be  negligible. 

The  temperature  of  the  water  bath  was  in- 
creased in  increments  of  about  2  °C  and  the  tis- 
sues were  allowed  to  equilibrate  with  the  bath 
water  prior  to  each  new  scan.    The  temperature 
of  the  bath  was  controlled  ±0.1  °C  and  experi- 
ments were  done  over  a  range  of  approximately 
22  to  42  °C. 

The  ultrasound  computerized  tomography  images 
for  the  acoustic  speeds  in  spleen,  kidney,  liver, 
spinal  cord  and  psoas  muscle  at  two  different 
temperatures  are  displayed  in  figure  4.  Since 
the  two  pictures  are  displayed  in  the  same  gray 
scale  any  increase  in  the  acoustic  speed  with 
temperature  is  seen  as  an  increase  in  the  bright- 
ness of  the  corresponding  regions  of  the  image. 
The  water  background  is  brighter  in  the  image  for 
33.2  °C  for  the  same  reason.    Figure  5  gives  a 
similar  picture  for  breast  tissues  at  tempera- 
tures 14  °C  and  41  °C.    Because  of  the  very  large 
temperature  change,  the  corresponding  changes  in 
the  acoustic  speeds  are  seen  strikingly  in  this 
picture.    The  water  background  goes  from  dark  to 
bright  and  the  image  of  salt  solution  gets  bright- 
er at  the  higher  temperature.    Images  of  the 
breast  fat  which  were  brighter  than  the  water 


229 


®0 


1  =  spleen 

2  =  kidney 

3  =  1 i  ver 

4  =  spinal  cord 

5  =  psoas  muscle 


Fig.  4.    Ultrasound  computerized  tomography  images 
of  acoustic  speed  distribution  in  various 
tissues,  in  vitro,  at  two  different  dif- 
ferent temperatures.    Same  gray  scale  used 
for  both  the  pictures.    Higher  acoustic 
speed  is  displayed  brighter.  Reconstruc- 
tion size  is  15  cm  square.    Tissues  were 
encased  in  rubber  finger  cots  to  prevent 
solutes  from  diffusing  into  or  out  of  the 
tissues.    Tissues  were  unfixed  and  main- 
tained at  10  to  14  °C  until  used  some  12 
to  24  hours  after  autopsy.  (Reproduced 
with  permission  from  Greenleaf,  et  al . , 
Proceedings  of  the  Vth  International 
Conference  on  Information  Processing  in 
Medical  Imaging  (in  press).) 

background  at  the  lower  temperature  become  darker 
at  the  higher  temperature  since  the  acoustic  speed 
in  the  breast  fat  has  decreased  with  increasing 
temperature.    The  dark  streak  from  fat  to  fat  in 
the  picture  for  41  °C  is  an  artifact  caused  by 
the  lens  effect  of  the  cyl indrically  shaped  tis- 
sue.   The  increase  in  the  acoustic  speed  in 


1  =  2.5  g/100  ml  salt 

sol ution 

2  =  breast  fat  with 

parenchyma 

3  =  breast  fat 


Fig.  5.    Ultrasound  computerized  tomography 

pictures  of  acoustic  speed  distributions 
in  breast  tissues,  in  vitro,  at  two  dif- 
ferent temperatures.    Reconstruction  size 
is  15  cm  square.    Fat  exhibits  acoustic 
speed  higher  than  water  at  14  °C  and 
lower  than  water  at  41  °C. 

parenchyma  with  temperature  is  also  seen  from 
this  picture. 

Results  of  the  statistical  analysis  of  the  re- 
construction images  giving  the  mean  and  standard 
deviation  of  the  acoustic  speeds  at  various  tem- 
peratures in  all  the  tissues  studied  are  given  in 
tables  1  and  2.    These  data  are  graphically  il- 
lustrated in  figures  6,  7,  and  8.    From  figure  6 
it  is  clear  that  the  tissues  liver,  psoas  muscle, 
spleen,  spinal  cord  and  kidney  have  higher  acous- 
tic speeds  than  water  and  their  temperature  varia- 
tions are  similar  to  that  of  water.    The  breast 
muscle  and  parenchymal  tissue  are  about  3  per- 
cent higher  in  acoustic  speed  than  normal  saline. 
Their  temperature  behavior  is  also  very  similar 


Table  1.    Temperature  variation  of  acoustic  speed  in  excised  human  breast  tissues. 

Temperature 


Tissue 

22.5  °C 

25.8  °C 

27.8  °C 

30.2  °C 

32.2  °C 

35.1  °C 

38.0  °C 

40.1  °C 

42.5  °C 

Salt  finger 
{2.5%) 

1516.73 
(2.1)b 

1523.2 
(2.1) 

1529 
(2.6) 

1533.6 
(2.1) 

1538.4 
(2.8) 

1544.8 
(3.1) 

1549.6 

(3) 

1552.8 
(4.2) 

1556.5 
(3.8) 

Fat  breast 

1480.7 
(2.5) 

1466.9 
(9.9) 

1472.1 
(9.8) 

1477.4 
(9.9) 

1478.1 
(11.4) 

1436.3 
(15.6) 

1435.6 
(18.4) 

1438.7 
(18.1) 

1441.8 
(17.5) 

Fat  with 
parenchyma 

1494.7 
(4.1) 

1487.5 
(5.5) 

1491.5 
(5.5) 

1497.8 
(5.7) 

1500.9 
(6.2) 

1471.3 
(9.7) 

1471.3 
(13.1) 

1474.2 

(13.4) 

1475.9 
(12.8) 

Parenchyma 

1539.4 
(4.5) 

1545.7 
(4.0) 

1551 
(4.6) 

1558.1 
(3.9) 

1562.1 
(4.1) 

1564.5 
(7.6) 

1564.3 
(6.4) 

1569.6 
(6.3) 

1571.9 
(6.4) 

Muscle 

1543. 1 
(5.2) 

1551.4 
(6.2) 

1554.2 
(6.5) 

1562.4 
(7.1) 

1565.5 
(6.4) 

1566.9 
(4.5) 

1570.7 
(6.9) 

1574.6 
(6.7) 

1579.5 
(5.4) 

Background 
(normal  saline) 

1504.0 
(3.2) 

1512.5 
(3.6) 

1519.2 
(2.8) 

1523.7 
(2.5) 

1528.4 
(2.4) 

1535.4 
(4.0) 

1539.1 
(2.9) 

15^1.9 
(3.0) 

1546.4 
(2.8) 

fvelocity,  m/s. 
Standard  deviation. 


230 


Table  2.    Temperature  variation  of  acoustic  speed  in  selected  excised  human  tissues. 


Temperature 


Tissue 

17.0  °C 

22.0  °C 

23.5  °C 

26.2  °C 

30.2  °C 

33.2  °C 

35.2  °C 

37.2  °C 

39.0  °C 

40.8  ° 

Liver 

1547.0^ 
(2.5)b 

1555.5 
(1.8) 

1563.1 
(3.0) 

1564.6 
(2.8) 

1571.1 
(2.5) 

1573.5 
(1.8) 

1575.3 
(2.5) 

1578.1 
(2.9) 

1580.0 
(2.2) 

1581.7 
(1.5) 

Kidney 

1508.5 
(4.3) 

1523.8 
(4.6) 

1536.1 
(2.1) 

1536.3 
(5.2) 

1545.2 
(2.7) 

1551.4 
(1.4) 

1555.8 
(1.8) 

1560.2 
(1.8) 

1562.6 
(1.2) 

1564.3 
(0.9) 

Spleen 

1528.0 
(1.8) 

1538.8 
(1.8) 

1544.3 
(1.8) 

1549.1 
(1.6) 

1556.4 
(2.0) 

1561.9 
(1.7) 

1546.0 
(2.1) 

1567.1 
(2.3) 

1569.3 
(2.6) 

1573.0 
(2.1) 

Psoas 
muscle 

1542.5 
(3.0) 

1459.5 
(3.0) 

1554.8 
(1.1) 

1560.2 
(1.7) 

1566.4 
(2.2) 

1571.6 
(1.8) 

1573.5 
(1.8) 

1575.6 
(1.1) 

1577.6 
(2.1) 

1580.3 
(1.8) 

Spinal 
cord 

1509.0 
(4.5) 

1523.0 
(4.6) 

1523.0 
(5.3) 

1526.0 
(3.0) 

1532.6 
(3.2) 

1538.0 
(2.6) 

1538.0 
(3.5) 

1542.4 
(3.3) 

1543.8 
(3.0) 

1-156.5 
(2.0) 

^Velocity,  m/s. 
Standard  deviation. 


Temperature,  °C 


Fig.  6.    Variation  of  velocity  of  sound  with  temper- 
ature in  various  tissues.    Ultrasound  com- 
puterized tomography  data.    Values  are 
averages  of  about  15  values  taken  in  the 
visually  defined  center  of  respective  tis- 
sue samples  within  the  image.    Bars  are 
±  RMSD. 

to  that  of  saline  or  water  as  seen  in  figure  7. 
However,  breast  fat  and  parenchymatous  fat  have 
acoustic  speeds  below  that  of  normal  saline  and 
show  a  complex  behavior  with  temperature  as  il- 
lustrated in  figure  8.    Initially  the  acoustic 
speed  increases  slowly  with  temperature  and  around 
^35  °C  decreases  markedly  and  then  increases  with  a 
'small  slope  of  about  1  meter  per  second  per  °C. 
Similar  decrease  but  much  less  pronounced  is  seen 
around  27  °C  also.    A  plot  of  the  variation  of  the 
temperature  coefficient  versus  temperature  for 
fat  and  muscle  shown  in  figure  9  illustrates  the 
strikingly  different  temperature  behavior  of  fat. 
This  complex  behavior  is  suggestive  of  possible 


•muscle 

□parenchyma 

isalt  solution,  2.5  g/100  ml  salt  solution 


Temperature,  °C 


Fig.  7.    Variation  of  acoustic  speed  within  pec- 
toral is  muscle  and  normal  parenchyma  of 
breast.    Values  obtained  in  the  manner 
described  in  figure  6. 

phase  transitions  just  below  the  body  tempera- 
ture (37  °C).    Hoyer  and  Nolle  [11]  studied  the 
behavior  of  nematic  and  cholesteric  liquid  crys- 
tals near  the  isotropic  to  liquid  crystal  transi- 
tion by  measuring  the  propagation  constants  of 
ultrasonic  waves.    Their  study  indicates  that  the 
acoustic  speed  goes  through  a  minimum  at  the 
temperature  of  the  phase  transition.    Dyro  and 
Edmonds  [12]  investigated  ultrasonic  dispersion 
in  cholesteryl  esters  and  their  results  show  that 
the  acoustic  speed  exhibits  markedly  different 
behavior  around  the  phase  transition.    Though  the 
fat  sample  in  the  present  study  is  not  a  homo- 
geneous, "clean",  one-component  system,  the  pos- 
sibility of  a  phase  transition  of  some  kind  can- 
not be  excluded.    More  detailed  studies  using 
different  physical  chemistry  techniques  such  as 
differential  scanning  calorimetry  may  shed  some 
light  into  the  anomalous  temperature  behavior  of 
the  breast  fat. 


231 


Temperature,  °C 


Fig.  8.    Variation  of  acoustic  speed  in  normal 

breast  fat.    Slope  of  alteration  in  acous- 
tic speed  with  temperature  is  negative  for 
fat  in  certain  regions  of  temperature. 
Vertical  shift  in  curves  is  apparently  due 
to  presence  of  parenchyma  which  has  tem- 
perature dependence  much  like  water  (see 
fig.  7). 

Alterations  of  temperature  within  various 
tissues  can  be  expected  under  various  inflammatory 
processes  such  as  infections.    It  is  also  well _ 
known  that  certain  carcinomas  are  associated  with 
a  high  temperature  due  to  their  increased  metabo- 
lism.   Knowledge  about  the  temperature  variations 
of  acoustic  speed  in  diseased  tissues  and  in 
normal  tissues  should  be  valuable  in  future 
studies  of  acoustic  speed  distributions  within 
patients  whose  tissues  are  undergoing  various 
disease  processes.    Computerized  ultrasound  tomo- 
graphy may  provide  the  possibility  of  quantita- 
tively measuring  acoustic  speed  in  breast  tissues 
in  vivo. 


Acknowledgments 

The  support  from  Dr.  Earl  H.  Wood,  Dr.  Erik  L. 
Ritman,  and  the  staff  of  the  Biodynamics  Research 
Unit,  Mayo  Clinic,  is  appreciated.    This  research 
was  supported  by  Grants  HL-07111,  HL-00170,  HL- 
00060,  RR-00007,  HL-04664,  NIH-HT-4-2904  from  the 
National  Institutes  of  Health,  United  States 
Public  Health  Service,  and  NCI-CB-64041  from  the 
National  Cancer  Institute. 

References 

[l]    Green,  P.  S.,  Schaefer,  L.  F.,  Jones,  E.  D.  , 
and  Suarez,  J.  R.,  A  New  Hi gh-Performance 
Ultrasonic  Camera,  in  Acoustical  Holography, 
Vol.  5,  pp.  493-503  (Plenum  Press,  New  York, 
1973). 

[2]    Jones,  J.  P.,  Impedi ography :    A  New  Ultra- 
sonic Technique  for  Non-Destructive  Testing 
and  Medical  Diagnosis,  in  Ultrasonics  Inter- 
national (1973),  Conference  Proceedings, 
p.  214  (PIC  Science  and  Technology  Press , 
U.K.,  1974). 

[3]    Kak,  A.  C.  and  Frye,  F.  J.,  Acoustic  Im- 
pedance Profiling:    An  Analytical  and 
Physical  Model  Study,  in  Ultrasound  Tissue 


-16l  1  I  I  I  I  I  I  I  I  L_ 

20         24         28  32         36  40 

Temperature,  °C 


Fig.  9.    Variation  of  temperature  coefficient  of 
acoustic  speed  with  temperature  in  pec- 
toral is  muscle  and  in  fat  from  breast. 
Negative  temperature  dependence  of  fat 
in  certain  regions  of  water  is  visualized 
by  plotting  slope  of  figure  8. 

Characterization,  M.  Linzer,  ed. ,  National 
Bureau  of  Standards  Spec.  Pub.  453,  p.  231 
(U.S.  Government  Printing  Office,  Washing- 
ton, D.C.,  1976). 

[4]    Greenleaf,  J.  F.  and  Johnson,  S.  A.,  Alge- 
braic Reconstruction  of  Spatial  Distributions 
of  Refractive  Index  and  Attenuation  in  Tis- 
sues from  Time-Of-Fl ight  and  Amplitude 
Profiles,  in  Ultrasonic  Tissue  Characteriza- 
tions ,  M.  Linzer,  ed. ,  National  Bureau  of 
Standards  Spec.  Pub.  453,  pp.  109-119  (U.S. 
Government  Printing  Office,  Washington, 
D.C.,  1976).  j 

[5]    Gordon,  R.  and  Herman,  G.  T. ,  Three- 
dimensional  reconstruction  from  projections: 
A  review  of  algorithms.  International  Re-  j 
view  of  Cytology  38,  111-151  (1974).  ' 

I 

[6]    Brooks,  R.  A.  and  Di  Chiro,  G.,  Principles  ! 
of  computer  assisted  tomography  (CAT)  in 
radiographic  and  radioisotopic  imaging, 
Phys.  Med.  Biol,  n,  689-732  (1976). 

[7]    Lakshminarayanan,  A.  V.,  Reconstruction 
from  Divergent  Ray  Data.    Department  of 
Computer  Science,  State  University  of  New 
York  at  Buffalo,  Technical  Report  92  (1975). 

[8]    Herman,  G.  T.,  Lakshminarayanan,  A.  V.,  and 
Naparstek,  A.,  Convolution  reconstruction 
techniques  for  divergent  beams,  Comput. 
Biol.  Med.  6,  259-271  (1976). 

[9]    Radon,  J.,  Ueber  die  Bestimmung  von  Funktion  'j 
durch  ihre  Integralwerte  Langs  gewisser 
Mannigfaltigkeiten,  Berichte  Saechsische 
Akademie  der  Wi ssenschaf ten  69,  262  (1917). 


232 


[10]    Millero,  F.  J.  and  Kubinski,  T.,  Speed  of 
sound  in  seawater  as  a  function  of  tempera- 
ture and  salinity  at  1  atmosphere,  J.  Acoust. 
Soc.  Am.  57,  312-319  (1975). 

[11]    Hoyer,  W.  A.  and  Nolle,  A.  W. ,  Behavior  of 
liquid  crystal  compounds  near  the  isotropic- 
anisotropic  transition,  J.  Chem.  Phys.  2A^, 
803-811  (1956). 

[12]    Dyro,  J.  F.  and  Edmonds,  P.  D. ,  Ultrasonic 
absorption  and  dispersion  in  cholesteryl 
esters,  MoT.  Cryst.  Liq.  Cryst.  25 ,  175- 
193  (1974T: 


233 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


HIGH  SPATIAL  RESOLUTION  ULTRASONIC  MEASUREMENT  TECHNIQUES 
FOR  CHARACTERIZATION  OF  STATIC  AND  MOVING  TISSUES 


Steven  A.  Johnson,  James  F.  Greenleaf,  and  B.  Rajagopalan 

Biodynamics  Research  Unit 
Department  of  Physiology  and  Biophysics 
Mayo  Foundation 
Rochester,  Minnesota    55901,  U.S.A. 


Robert  C.  Bahn 

Department  of  Pathology 
Mayo  Clinic  and  Mayo  Foundation 
Rochester,  Minnesota    55901,  U.S.A. 


Brent  Baxter 

Department  of  Computer  Science 
University  of  Utah 
Salt  Lake  City,  Utah    84112,  U.S.A. 


Douglas  Christensen 

Departments  of  Bioengineering  and  Electrical  Engineering 
University  of  Utah 
Salt  Lake  City,  Utah    84112,  U.S.A. 


Clinical  and  pathology-based  arguments  are  presented  for  the  need  for  higher  resolu- 
tion ultrasound  images.    The  theoretical  foundation  and  experimental  characteristics 
of  a  high  resolution,  sampled  aperture,  reflection  technique  termed  "synthetic  focus" 
imaging  are  given.    It  is  shown  by  theory  and  simulation  that  such  synthetic  focus 
images  may  be  corrected  for  attenuation  and  refraction  effects  and  thereby  approach 
one-half  wavelength  resolution.    The  similarity  between  synthetic  focus  and  seismic 
migration  techniques  is  discussed.    An  example  of  a  high  resolution,  seismic  process- 
ed (i.e.,  migrated)  image  obtained  from  real  data  at  medical  ultrasound  frequencies 
is  shown.    Synthetic  focus  imaging  theory  is  extended  to  moving  coordinate  systems 
and  the  effect  of  Doppler  shift  effects  on  echo  pulse  shape  is  discussed.    A  general- 
ized wide  aperture  Doppler  imaging  theory  is  presented  which  suggests  further  improve- 
ments in  signal-to-noise  ratio,  spatial  resolution  and  flow  velocity  over  narrow 
aperture  systems  is  possible.    A  new  computed  tomographic  flow  measurement  and  recon- 
struction technique  based  on  transmission  measurements  is  presented.    This  technique 
permits  imaging  the  three  flow  velocity  components  and  temperature  of  homogeneous 
fluids  in  a  three-dimensional  domain. 


Key  words:    Aperture  synthesis;  computed  tomography;  Doppler;  fluid  flow;  high 
resolution;  reconstruction;  temperature  reconstruction;  ultrasound. 


1.  Introduction 

This  paper  presents  theoretical  and  experi- 
mental evidence  that  useful  measurements  of  tis- 
sue characteristics  with  limiting  spatial  re- 
solutions of  about  one-half  wavelength  are  pos- 
sible in  static  and  moving  media  (e.g. ,  tissues). 


This  paper  has  three  objectives:    First,  to 
present  the  case  for  striving  for  higher  re- 
solution images  of  tissue  parameters,  second, 
to  describe  the  theoretical  and  experimental 
techniques  whereby  such  high  resolution  images 
may  be  obtained  and,  third,  to  report  progress 
toward  obtaining  such  high  resolution  images. 


235 


2.    The  Case  for  Seeking  Higher 
Resolution  Images 

A.    Higher  Resolution  Images  of  Static  Tissues 

The  first  goal  or  purpose  of  this  paper  is  in 
part  philosophical,  yet  is  demonstrable,  and  calls 
attention  to  an  important  area  of  investigation 
(improved  resolution)  where  future  research  prog- 
ress can  be  made.    It  can  be  argued  that  reliable 
tissue  characterization  requires  attention  to  at 
least  three  independent  requirements:  First, 
choosing  an  adequate  set  of  tissue  dependent  acous- 
tic parameters  to  be  measured;  second,  measuring 
the  parameters  with  accuracy  and  precision  in  some 
set  of  regions  in  the  tissue;  and  third,  making 
this  measurement  region  as  small  as  possible  and 
as  densely  arranged  or  packed  as  possible  (i.e., 
forming  an  image). 

The  third  requirement  is  not  always  necessary 
to  gain  some  information  on  tissue  type  but  its 
inclusion  improves  the  certainty  of  characteriza- 
tion.   Certainly,  it  is  clear  that  an  early  diag- 
nosis of  a  disease  state  (such  as  cancer)  is  en- 
hanced by  detection  of  smaller  masses  made  pos- 
sible by  higher  resolution  and  better  tissue 
identification.     In  some  cases  the  tissue  type  may 
well  be  primarily  determined  by  the  morphology  and 
pattern  of  its  structure  and  the  change  in  pattern 
of  the  surrounding  tissue  [l,2]i.    This  has  its 
analog  in  microscopic  pathologic  techniques  where 
both  stain  (acoustic  tissue  characterization 
parameter  is  its  analog)  and  pattern  are  important 
in  tissue  typing  or  characterization. 

In  the  long  range,  the  improvement  of  resolu- 
tion may  be  the  most  important  objective  for 
future  research  progress.    At  present,  the  pri- 
mary attempt  to  detect  small  focal  processes  such 
as  cancers,  abscesses,  infarcts  of  less  than  0.5 
to  1  cm  in  diameter  is  difficult  and  may  not  be 
justified  economically.    However,  often  more 
fundamental  than  the  focal  lesion  itself,  regard- 
less of  its  size,  is  the  state  of  the  tissue  sur- 
rounding circumscribed  lesions.    The  pattern  of 
the  adjacent  tissue  plays  a  crucial  role  in  the 
identification  of  specific  diseases  by  supplying 
the  physician  with  information  concerning  the 
local  context  of  the  process  under  consideration. 

Compared  to  tumor  nodules,  the  structures  of 
the  surrounding  "normal"  tissues  are  relatively 
delicate.    Their  patterns  are  essentially  deter- 
mined by  the  dimensions  of  the  fibrous  and  vascu- 
lar framework  of  an  organ.    It  would  be  desirable 
to  image  such  delicate  patterns  associated  with 
the  tertiary  branching  of  major  arteries  of  the 
heart,  brain,  kidneys,  and  lungs,  the  bronchial 
tree,  the  biliary  ducts  of  the  liver  and  the 
ductular  system  of  the  breast.    Since  vascular 
or  ductal  systems  at  the  level  of  tertiary  branch- 
ing involve  tubular  structures  approximately  0.5 
to  1.0  mm  in  diameter,  this  objective  fixes  the 
upper  limit  of  useful  picture  element  size  at 
about  0.25  to  0.5  mm.    From  a  practical  point  of 
view,  this  level  of  resolution  would  provide  at 
least  20  resolutable  picture  elements  (pixels) 
over  an  area  2.5  millimeters  in  diameter.  Such 
an  area  would  correspond  roughly  to  the  cross- 
sectional  area  of  the  lumen  and  portions  of  the 
wall  of  a  coronary  artery. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


In  this  manner,  imaging  studies  can  eventually 
become  focused  upon  pathogenesis,  prevention,  and 
early  detection  of  disease  rather  than  upon  the 
diagnosis  of  gross  advanced  lesions. 

B.    The  Case  for  Seeking  High  Resolution 
Images  of  Moving  Tissues 

The  velocity  of  moving  tissue  may  be  obtained 
by  two  general  methods.    First,  by  a  differential 
method  in  which  velocity  is  computed  from  images 
of  the  location  of  the  tissue  at  two  separate 
times  separated  by  a  short  time  interval  (this 
could  be  done  by  comparing  two  separate  real-time 
B-scans  of  the  heart,  for  example),  and  second, 
by  direct  methods  which  make  use  of  alterations 
in  frequency  or  phase  in  the  received  ultrasound 
signals.    Examples  of  images  produced  by  this 
latter  method  are  the  familiar  pulsed  Doppler  B- 
scan  technique  [3]  and  the  new  fluid  velocity 
vector  reconstruction  technique  developed  by 
Johnson  et  a1 .  [4]. 

Images  produced  by  the  first  or  differential 
method  have  application  in  understanding  moving 
solid  tissues  such  as  the  heart  and  circulatory 
system  and  their  components  (e.g. ,  valves,  ves- 
sel walls,  etc . ) .    Any  improvement  in  the  spatial 
and  temporal  resolution  of  methods  for  imaging 
these  moving  structures  would  produce  a  corre- 
sponding improvement  in  imaging  the  velocity  of 
these  structures  at  high  spatial  resolution.  The 
effort  which  has  been  and  now  is  being  made  in 
the  field  of  x-ray  angiography  [5]  is  an  indica- 
tion of  the  potential  contribution  which  such 
high  spatial  and  temporal  ultrasound  imaging 
methods  could  make  to  health  care. 

Images  produced  by  the  direct  velocity  imaging 
techniques  (based  on  frequency  or  phase  shift 
techniques)  have  application  in  measuring  the 
velocity  of  fluid  produced  or  modified  by  tissues 
(e.g . ,  air  in  the  respiratory  system,  urine,  etc. ) 
or  the  velocity  of  fluid  tissues  (e.g. ,  the  blood). 
The  well-known  Doppler  technique  (a  reflection 
method)  measures  the  velocity  of  scattering  cen- 
ters (e.g. ,  the  red  and  white  cells)  in  the  blood. 
The  new  velocity  vector  reconstruction  method  (a 
transmission  method)  measures  the  velocity  of  the 
carrier  (i.e.,  the  plasma)  of  the  scattering  cen- 
ters or  a  combination  of  the  velocity  of  the  two 
constitutive  components.    Thus,  the  two  methods 
are  complementary.    The  ability  to  obtain  images 
at  high  spatial  and  temporal  resolution  of  blood 
flows  promises  to  be  of  great  value  for  several 
reasons.    First,  it  may  be  argued  that  since  the 
x-ray  angiographic  method  is  of  proven  worth, 
then  any  competitive  ultrasound  techniques  would 
also  be  valuable.    Second,  the  new  pulsed  Doppler 
systems  now  in  development  or  use,  although  of 
limited  to  moderate  imaging  capability,  are  prov- 
ing to  be  of  great  value  [6].    For  example,  the 
SRI-built  carotid  scanner,  now  undergoing  clinical 
trials  at  Mayo,  is  proving  the  value  of  its  Doppler 
flow  measuring  features  (the  scanner  has  real- 
time B-scan  capability  also)  [7].    Third,  ultra- 
sound procedures  are  noninvasive  and  have  no 
demonstrable  cumulative  toxic  effect  when  power 
and  energy  levels  are  controlled.    Fourth,  ultra- 
sound procedures  are  painless  and  usually  can  be 
completed  quickly. 


236 


3.    Imaging  Methods  for  Static  Media 

A  new  high  resolution  reflection  imaging  mode 
has  been  developed  in  the  Mayo  laboratory  to 
image  or  characterize  tissue  structures  in  the 
body  where  a  full  360  degree  data  collection 
geometry  is  precluded  by  the  ribs,  the  spine,  and 
other  body  structures  [7,8].    This  new  mode  is 
termed  synthetic  focus  imaging  [7]. 

The  mathematics  for  synthetic  aperture  imaging 
may  be  derived  by  reference  to  figure  1.  Let 
define  a  measure  of  the  probability  for  acoustic 
amplitude  scattering  or  reflection  toward  the  de- 
tector aperture  from  picture  element  k  when  il- 
lumination from  each  transmitter  element  in  the 


water  tank    Gmk  9jk 


\ 

n 

:\ 

1 

\f 

k 

1 

— 

— 

Tik 

11 

(M,M) 


(M,0) 


transducer  array 


Vijm 


12  3 
i(j,k,m)-l 


d  = 
i  (j,k,m)  +  l 


Fig.  1. 


Geometry  for  derivated  of  equations  for 
ultrasonic  imaging  by  generalized 
synthetic  aperture  methods.    A  transducer 
array  with  N  elements  is  in  acoustic  con- 
tact with  a  water  tank.    Positions  in  the 
tank  are  described  by  coordinate  system 
XY.    A  typical  ray  path  is  shown  for 
energy  transmitted  by  the  mth  transducer 
and  scattered  by  picture  element  k  into 
the  jth  transducer.    Here  rji^  and  r^k 
are  the  distances  from  j  to  k  and  from 
m  to  k,  respectively.    Angle  6j|<  is  mea- 
sured from  rj|<;  to  the  Y  axis  counter- 
clockwise.   V-jjni  is  the  voltage  sample 
from  the  jth  transducer  at  time  after 
transmission  of  the  signal  by  the  mth 
transducer.    Here  s  and  d  are  start  and 
end  times  of  a  time  interval  2w  which 
contains  the  echo  from  k  at  its  midpoint. 
Time  L  is  greater  than  the  maximum  time 
associated  with  all  paths  defined  by  j, 
m,  and  k.    (Reproduced  with  permission 
from  [7].) 


array  aperture,  measured  at  picture  element  k, 
is  normalized.    The  probability  for  energy  scat- 
tering is  the  square  of  P^.    For  the  case  of  a 
very  narrow  pulse  of  ultrasonic  energy,  P|^  is 


given  by 


N 

E 

m=l 


N 
j  =  l 


t+w 

E 

i=t-v\ 


^i,j,mRj,kTk,mKt-l  (D 


where  Vi,j^ni  is  the  ith  voltage  sample  from  the 
j'^n  array  element  when  the  mtn  array  element  was 
used  as  the  transmitter.    Rj^|<  is  the  receiver- 
range  azimuth  illumination  normalization  factor 
between  the  k^n  picture  element  and  the  jth  array 
element.  Thus,  Rj^|^  is  independent  of  time,  but 
is  an  approximate  function  of  some  power  of  the 
magnitude  of  rj^|<  (where  rj^^  is  the  vector  from 
array  element  j  to  picture  e'l_ement  k)  and  of  the 
cosine  of  the  angle  between  fj  ^  and  n-;  (the  unit 
vector  normal  to  array  element'j).    Thus,  R,-  i, 
may  be  written  ' 


exp 


(/a(s)ds) 


-   j  " 

q 

/ds 

n.  •  S,  . 

'-k 

-1 


(2) 


where  a  is  the  attenuation  coefficient  along  the 
ray  path  connecting  k  and  j,  q  is  a  rational  num- 
ber from  0.5  for  cylindral  waves  to  1.0  for 
spherical  waves,  and  is  the  unit  tanget  vec- 

tor at  j  of  the  ray  which  passes  from  k  to  j. 
T)^  u,  is  the  corresponding  transmitter  function 
belween  the  k^h  pixel  and  mth  array  (transmitter) 
element.    Thus,       |^  and  T|^^rn  affect  the  energy 
normalizations  referred  to  in  the  definition  of 
P|<.    N  is  the  number  of  array  elements.    The  fac- 
tor K^_-j  is  a  kernel  function  for  performing  con- 
volution or  cross  correlation  operations  on  the 
raw  data  V|  h         If  V-j         is  of  the  appropriate 
pre-transmi{ted  form  or 'post-received  processed 
form,  then  K^.-j  may  be  of  the  form  of  a  delta 
function  6q        =  6t,i.    The  limits  t  ±  w  are 
related  to  {he  transmitter  pulsewidth  t^  which  is 
less  than  some  arbitrary  bracketing  interval  tB 
which  contains  the  transmitted  pulse  of  width  t^. 
Then, 


/    k  j 


ds 


At 


tw/2At 


(3) 


(4) 


where  At  is  the  sampling  interval  time  between 
successive  samples  in  the  digitized  signal,  and  u 
is  the  effective  velocity  of  sound  in  the  insoni- 
fied  medium.    The  two  line  integrals  are  taken 
along  the  rays  connecting  transmitter  m  to  k  and 
k  to  receiver  j  respectively. 

An  important  modification  to  the  synthetic 
focus  equations  above  would  change  the  sum  over  i 
to  allow  for  refraction  or  bending  of  acoustic 
rays.    The  effect  of  this  bending  is  manifest  in 
shifts  in  the  values  of  s  and  d  compared  to  that 
obtained  by  computing  the  round  trip  pulse  time  in 
a  homogeneous  medium.    This  effect  is  illustrated 
in  figure  2.    The  value  of  t  would  be  replaced  by 
the  correct  round  trip  time  computed  using  curved 
rays  obtained    from  ray  tracing  techniques  [9,10]. 
The  coordinate  system  set  up  by  the  intersection 
of  rays  and  wavefronts  and  the  corresponsi ng 
slight  shift  in  round  trip  (a  few  percent)  will 
allow  a  low  spatial  resolution  perturbation  cor- 
rection to  be  made  to  the  basic  homogeneous  media 
round  trip  time  equations. 

Another  useful  modification  to  the  basic 
synthetic-focusing  scheme  presented  above  would 


237 


Fig.  2.    Orthogonal  coordinate  systems  for  describ- 
ing ray  paths  from  transducer  elements  in 
constant  (left)  and  variable  (right)  re- 
fractive index  media.    On  the  left  the  ray 
paths  between  each  transducer  element  and 
receiver  element  are  segments  of  straight 
lines.    The  lines  of  equal  time  of  propa- 
gation are  circles.    On  the  right  the  ray 
path  and  equal  time  lines  are  from  trans- 
mitter element  m  to  scattering  site  k  and 
the  time  tj|^  from  k  to  receiver  element  j 
is  shown  as  a  heavy  line.  (Reproduced 
with  permission  from  [9]). 

be  to  replace  ^■\,j,m  with  the  output  of  a  hard- 
ware or  software  correlator,  allowing  transmission 
not  only  of  pulses  of  energy,  but  arbitrary  wave- 
forms as  well.    This  may  be  a  faster  method  for 
obtaining  the  optimum  resolution  waveforms  than 


performing  the  inner  products  in  software  using 
the  kernel  Kt-i  as  indicated  by  equation  (1). 

Figure  3  shows  the  similarities  between  the  x- 
ray  computer  assisted  tomography  and  synthetic 
focus  ultrasound  algorithms.    It  is  seen  that  back 
projection  in  both  instances  can  be  thought  of  as 
a  line  integral  in  data  space: 

1)  In  the  x-ray  case  the  reconstructed  value 
at  each  pixel  in  the  tomogram  is  obtained  by  in- 
tegration along  a  unique  path  in  the  convolved 
data  space.    The  convolved  data  space  is  composed 
of  the  collected  projection  profiles  convolved 
with  a  kernel  function.    The  curved  path  associat- 
ed with  each  pixel  is  a  sine  wave  modified  in 
phase  and  amplitude  for  the  special  case  of 
parallel  ray  projections,  hence,  the  name  "sino- 
gram"  space  for  this  representation  of  the  data. 

2)  In  the  ultrasound  synthetic  focus  method, 
the  image  value  at  each  pixel  in  the  tomogram  is 
obtained  by  integration  along  a  unique  path,  also 
in  a  convolved  data  space.    The  values  in  the  con- 
volved data  space  can  be  obtained  by  either  trans- 
mitting the  required  waveform  or  by  digital  proc- 
essing or  equalization  of  the  received  data.  In 
the  case  of  a  homogeneous  material  in  the  object 
space,  e.g. ,  water,  the  set  of  echoes  from  each 
receiver  element  for  a  fixed  transmitter  position 
falls  on  a  hyperbolic  curve  in  data  space.  The 
synthetic  focus  algorithm  reconstructs  the  scat- 
tering amplitude  at  each  pixel  by  integration 
along  the  hyperbolic  curve  which  is  unique  to  that 
pixel . 

Back  projection,  as  just  described,  may  be 
termed  "pixel  driven"  because  the  value  at  each 
pixel  is  computed  to  completion  before  a  value  is 


0  a  ->  TT  21T 


Fig.  3.     Illustration  of  the  similarity  between  "back  projection"  reconstruction  in  the  x-ray 
case  and  echo  mode  synthetic  focus  image  formation  with  ultrasound. 
Left  panel  -  Reflection  Ultrasound  Data  Space  -  Pixel  driven  synthetic  focus  algorithm 
generates  a  unique  hyperbolic  curve  along  which  "back  projection"  algorithm  performs 
a  line  integral  for  the  data  value  at  each  pixel  value. 

Right  panel  -  Transmission  X-ray  Data  Space  -  Pixel  driven  convolution  algorithm 
generates  unique  sine  wave  like  curve  along  which  "back  projection"  algorithm  per- 
forms a  line  integral  for  the  data  value  at  each  pixel  value. 


238 


assigned  to  a  new  pixel.  The  line  integration 
along  curves  in  data  space  is  assigned  to  each 
pixel  in  this  mode  of  computation. 

An  alternative  conceptualization  of  back  pro- 
jection is  possible  in  which  each  data  point  in 
data  space  is  used  to  influence  the  value  of  all 
pixels  in  the  image  space  before  a  new  data  point 
is  processed.    This  formulation  may  be  termed 
"profile  driven"  back  projection  for  the  reasons 
just  given,  or  termed  "smear  and  add"  back  pro- 
jection because  the  pixels  which  are  influenced 
most  by  each  data  point  are  located  on  or  near  a 
curve  in  image  space.    The  "smear"  may  refer  to 
convolution.    In  the  x-ray  case,  the  "smear  and 
add"  occurs  back  along  the  original  x-ray  path 
which  produced  the  data  value  at  the  detector. 
This  is  a  very  strong  intuitive  reason  for  the 
term  "back  projection".    In  the  ultrasound  case, 
the  back  projection  in  the  image  space  occurs 
along  an  ellipse  (not  a  hyperbole)  for  each  data 
space.    The  ellipse,  of  course,  represents  the 
locus  of  points  of  equal  transmitter-receiver 
pair  round  trip  time.    The  superposition  of  the 
set  of  values  along  all  ellipses  for  all  such 
transducer    pairs  constitutes  the  final  ultra- 
sound scatter  amplitude  image  in  an  analogous 
manner  to  the  formation  of  an  x-ray  convolution 
algorithm  image  by  the  sum  of  all  back  projected 
convolved  projections.    This  viewpoint  is  further 
explained  in  figure  4. 


r-i  source 

Vl.5(t) 


data  space 


14,11 


(t) 


-t 


ul trasound 
transducer  array 


A.    Diffraction  Considerations 

Although  no  diffraction  theory  has  been  used 
in  the  derivation  of  the  synthetic  focus  equations, 
it  is  still  true  that  images  obtained  from  their 
application  are  very  nearly  diffraction  limited 
(at  least  qualitatively  and  quantitatively  in 
some  cases).    The  reason  for  this,  perhaps  surpris- 
ing, result  lies  in  the  use  of  an  extremum  ray 
path  time  between  the  scattering  center  (image 
point)  and  transducer  element  in  the  synthetic 
focus  method.    These  ray  paths  are  often  identical 
to,  or  very  nearly  equal  to,  the  characteristic 
solution  lines  to  the  wave  equation  (the  wave 
equation  includes  diffraction  effects).  Thus, 
although  some  energy  travels  by  diffracted  paths, 
a  large  fraction  usually  will  travel  by  the  ex- 
tremum paths  used  by  the  synthetic  focus  method. 

B.    Comparison  With  Seismic  Methods 

Many  similarities  exist  between  the  synthetic 
focus  ultrasound  algorithm  developed  by  Johnson 
et  al  ■  [7]  and  classical  seismic  migration  methods 
[ 11 ] .    One  similarity  is  in  the  method  of  collect- 
ing data.    Both  migration  and  synthetic  focus 
methods  use  the  concept  of  separate  source  or 
transmitting  points  and  geophone  or  receiving 
points  respectively.    This  is  illustrated  in 
figure  5  for  the  case  of  a  linear  geophone  and 
circular  synthetic  focus  ultrasound  arrays. 


source  circle 


ray  source 


projection 
profile 


convol ved 

projection 

profile 


Fig.  4.    Illustration  of  the  ultrasound  analog  of  back  projection  of  a  convolved 
projection  profile. 


The  left  figure  shows  the  signal  voltage       5(t)  received  on  element  5 
when  transmitting  from  element  1  as  a  function  of  time.    Also  shown  is  the 
voltage  V14  ii(t)  for  receiver  11  and  transmitter  14.    For  simplicity,  it 
is  assumed  {hat  the  object  space  contains  only  one  scattering  point  Q. 
The  data  Vi,j(t)  is  assumed  to  be  ready  for  back  projection.    The  right 
figure  shows  an  x-ray  source  and  object  with  a  dense  point  at  P  which 
produces  the  sharply  peaked  projection  profile.    Reconstruction  of  the 
object  is  achieved  by  back  projection  of  the  convolved  profiles  for  all 
source  positions  around  360°.    In  the  ultrasound  case,  the  receiver  echo 
data  is  projected  along  ellipses  in  the  object  space  (note  that  this  is 
equivalent  to  summing  along  a  set  of  hyperbolae  in  the  data  space).  As 
in  the  x-ray  case,  back  projection  of  non-negative  unconvolved  raw  projec- 
tion profile  data  does  not  give  optimum  resolution  but  instead  gives  the 
true  object  convolved  with  a  blurring  function.    The  effect  of  the  blurring 
function  may  be  nearly  removed  (deconvolved)  if  the  projection  profiles  are 
first  convolved  with  a  deblurring  kernel.    Such  kernels  usually  have  a 
strong  central  maxima  with  symmetric  heavily  damped  bipolar  side  lobes. 
In  the  ultrasound  case  maximum  resolution  (the  Rayleigh  limit  and  even  much 
better)  is  achieved  when  the  impulse  response  of  a  single  scatterer  is  made 
to  look  like  such  a  kernel  in  the  data  space. 


239 


s 

? 

'f=2 


.5d 


transmission  mode 
receiver  array 


f  J 

I  I 


reflection  mode 
receiver 


y  [— y 
'  I 

-1  9+1 


1 


=  -1 


DATA 
COLLECTION 
GEOMETRY 


-3 


-2  -1  0  +1  +2 
—I — I — I — I — 


'^^^  si^':^a'%  gc  gd  9e 


-4 


+4 

^  3 

■ 

.d  ^ 

0 

-  •  c 

.  b 

•  a 

DATA 
DISPLAY 
DOMAIN 


0  12  3  4 
source  axis   


prof i 1 e 

(common  source] 


►common  receiver 
prof i 1 e 

^ommon  midpoint 
gather 

seismic  section 
(common  offset) 


DATA 
COLLECTION 
SCHEMES 


"IL    w  0  1 
2  2 
source  angle  axis  — 

J  circular  profile 
I  (common  source) 


■common  receiver 
circular  profile 


(^common  midangle 
gather 

acoustic 

circular  section 


Fig.  5.    Illustration  of  the  similarities  between 
classical  seismic  data  collection  with 
linear  geophone  arrays  and  ultrasound 
imaging  with  circular  transducer  arrays. 
The  left  most  column  of  figures  illustrates 
the  major  parameters  used  to  describe  seis- 
mic data  collection  in  one-dimensional, 
i.e.,  linear,  arrangements  of  source 
positions  (shot  point  =  s)  and  receivers 
(geophone  =  g).    The  right  most  column  of 
figures  illustrates  how  the  ci rcumgerence 
of  a  circular  arrangement  of  sources  and 
receivers  may  be  described  in  terms  of  the 
analogous  parameters  in  the  left  column. 
The  shot  position  s  is  replaced  by  source 
angle  65  and  geophone  position  g  is 
replaced  by  the  receiver  angle  eg. 

In  the  top  most  left  figure,  typical  source 
s  is  located  at  position  -3  and  typical 
geophone  g^  is  located  at  position  +1 
(arbitrary  units).    The  mid  point  of  the 
(s,g(j)  pair  is  at  position  -1  and  is  de- 
fined as  the  mid  point  y  (i.e.,  y  =  -1). 
The  separation  of  s  and  gj  from  the  mid- 
point is  -2  and  +2  respectively  and  is  de- 
fined as  the  offset  f  (f  for  "offset" ) . 
Note  f  is  positive  when  g  is  on  the  posi- 
tive side  of  the  midpoint. 

Also  shown  is  a  typical  seismic  data  col- 


lection geometry  with  a  source  (dynamite 
blast)  located  at  -3  and  geophones  at  -2, 
-1,  0,  1,  2.    This  particular  source- 
receiver  combination  is  plotted  as  points 
a,  b,  c,  d,  e  on  the  g-s  plane.    Point  d 
at  (-3,1)  in  the  g-s  plane,  represents 
time  history  data  for  a  source  at  -3  and 
a  geophone  at  +1.    The  midpoint  and  offset 
coordinates  (-1,2)  of  data  point  d  can  be 
found  from  its  projection  onto  the  midpoint 
and  offset  axes  respectively.    The  direc- 
tions of  straight  line  sets  of  data  points 
at  angles  of  0,  45°,  90°,  and  135°  to  the 
s-axis  have  important  properties  and  are 
given  characteristic  names.    A  vertical 
line  is  a  prof i 1 e  and  corresponds  to  a  data 
point  with  a  common  source  (the  usual  mode 
of  seismic  collection).    A  horizontal  line 
is  a  common  receiver  profile  (rarely  used), 
a  line  at  45°  is  a  seismic  section  (common 
offset)  and  is  often  obtained  by  rearrange- 
ment of  multiple  profiles.    A  seismic  sec- 
tion is  also  commonly  generated  by  towing 
a  geophone  at  a  fixed  separation  behind  a 
source  in  the  ocean.    A  line  at  135°  is 
called  a  common  midpoint  gather. 

The  top  most  drawing  on  the  right  shows  how 
the  concepts  described  in  the  left  column 
can  be  applied  to  circular  geometries.  The 


240 


Fig.  5.  (continued) 


linear  parameters  s,  g, 
angular  analogues,  9$, 
respectively. 


and  f  have  their 
Gw,  and  Gf 


The  s-g  plane,  which  extends  to  infinity 
on  all  sides,  is  replaced  by  a  finite  and 
bounded  Qs-Qg  plane  (bounded  by  ±  tt).  The 
Gs-eq  plane  provides  a  convenient  means 
for  displaying  the  data  collected  for  x-ray 
or  ultrasound  transmission  computed  recon- 
struction tomography.     In  the  top  right 
figure,  a  circular  receiving  array  covering 
180°  is  shown  opposite  the  source  s.  Trans- 
mission data  collected  by  this  array  is 
contained  in  the  trapazoids  DEFG  and  HIJK. 


Reflection  data  for  s  coincident  with  g  is 
found  on  the  line  AOB.    Reflection  data  for 
the  receiver  offset  from  the  source  line 
AOB.    Reflection  data  for  the  receiver  off- 
set from  the  source  as  shown  is  found  along 
lines  mnp  and  qu.    (This  is  one  line  mnpqr 
when  the  periodic  nature  of  the  data  space 
is  considered). 

The  direction  of  straight  line  data  sets  in 
the  Gs)  Gg  plane  can  be  given  analogous 
names  as  those  defined  in  the  s-g  plane. 
These  data  sets  are  shown  at  the  bottom  of 
the  right  column  (i.e.,  circular  profile, 
common  receiver  circular  profile,  common 
midangle  gather,  and  acoustic  circular 
sections) . 


typical 
signal 


reflection,  scatter  and 
Doppler  Scatter  space 
above  EAT  surface 


earliest  arrival 
time  (EAT)  surface 


transmi  ssion 
mode  array 


domain  null  space 
beneath  EAT  surface 


Fig.  6.    Illustration  of  the  mathematical  proper- 
ties of  acoustic  data  collected  from 
source-receiver  pairs  on  a  circular 
boundary.    The  time  dependent  voltage 
of  the  signal  received  by  receiver  g  is 
plotted  as  an  amplitude  (fourth  amplitude 
dimension  not  shown)  v_s_. ,  Gs,  Gg,  and 
time  t.    The  coordinates  9$,  Gg  are 
source  and  receiver  positions  as  defined 
by  the  drawing  at  the  right.    For  any  two 
points  s  and  g  on  the  circumference,  there 
exists  an  earliest  arrival  time  (EAT)  for 
acoustic  energy  to  propagate  from  s  to  g. 
This  time  is  zero  if  s  is  coincident  with 
g  but  is  maximum  (or  near  maximum  for  non- 
homogenous  substances)  when  s  and  g  are  on 
opposite  ends  of  a  diameter.    Thus,  for 
times  t  1  ess  than  EAT,  the  values  in  the 
range  (not  the  domain)  of  (Gs,  Gg,  t)  are 

The  similarity  between  linear  array  seismic 
methods  and  synthetic  focus  medical  ultrasound 
methods  exists  for  both  linear  and  circular 
scanning  geometries.    Circular  geometries  are 
found  in  x-ray  computed  tomographic  reconstruc- 
tion instruments  and  in  their  ultrasound  counter- 
parts developed  at  Mayo  Clinic  by  Greenleaf  and 
Johnson  [121.    This  similarity  is  explored  and 
developed  in  figure  5  and  its  legend.    Many  of 
the  important  features  of  transmission  mode 
reconstruction  imaging  such  as  "sinogram"  space, 
or  transformation  from  fan  beam  to  parallel  beam 
geometries,  can  be  analyzed  with  the  aid  of  the 
well-known  seismic  s-g  plane  data  representation 
(s-g  means  source  and  geophone  pair  location  for 
each  data  record). 


zero  (no  signal  has  as  yet  arrived).  Thus, 
the  EAT  value  assigned  to  each  pair  (Gs, 
Gg)  defines  a  surface  below  which  all 
values  in  the  same  range  are  zero.  This 
surface  is  called  the  EAT  surface.  The 
region  above  the  0-z  axis  (2j_e. ,  the 
9s  ~  9q  line)  provides  maximum  time  separa- 
tion of  echoes  from  most  neighboring  points 
and  provides  maximum  Doppler  shifts  for 
fluid  flows.    The  region  above  areas  abed 
and  ehij  corresponds  to  receiver  positions 
nearly  opposite  from  the  source  and  are 
used  for  computed  tomographic  (i.e.,  C.T.) 
transmission  data  collection.    A  maximum 
arrival  time  (MAT)  surface  (not  shown) 
also  can  be  drawn  for  which  all  points 
above  this  surface  correspond  to  multiple 
reflections  or  multiple  scattering  events, 
or  both. 

The  nature  of  the  s-g  data  domain  representa- 
tion in  circular  geometries  can  be  extended  to 
include  the  time  variable.    An  explanation  of  the 
features  of  this  three-dimensional  data  domain  is 
given  in  figure  6.    There  it  is  shown  that  the 
domain  can  be  partitioned  into  two  regions  by  a 
surface  which  corresponds  to  the  earliest  arrival 
time  of  an  acoustic  pulse  or  signal.    The  distance 
between  a  point  P  on  this  surface  and  the  s-g 
plane  corresponds  (nearly  inversely  proportional- 
ly) to  the  "merit"  of  the  echo  spatial  resolution 
information  contained  in  the  signal  associated 
with  the  point  P  (i.e.,  a  receiver  opposite  a 
source  can  have  no  or  little  echo  information). 

The  synthetic  focus  method  reported  by  Johnson 
et  al .  [7],  according  to  the  seismic  nomenclature. 


241 


collects  a  profile  for  each  transmitter  position. 
The  side  looking  radars  or  side  looking  synthetic 
aperture  radars  can  be  classified  in  seismic  terms 
as  producing  one  zero-offset  section  per  flight 
path.    It  is  well-known  that  superior  focused 
(i.e.,  "migrated")  seismic  images  can  be  obtained 
by  the  use  of  multiple  profiles  rather  than  by  the 
use  of  one  zero-offset  section  [11].    Thus,  our 
synthetic  focus  method  should  produce  images 
superior  to  those  produced  by  literal  adaptation 
of  side  looking  radar  or  synthetic  aperture  tech- 
niques to  medical  imaging. 

Also,  in  radar  technology,  refraction  (i.e., 
normal  moveout  corrections)  are  not  usually,  if 
ever,  made  (but  should  be  for  medical  imaging). 

4.     Imaging  in  Moving  Media 

A.    General  Considerations 

Imaging  in  the  presence  of  moving  media  re- 
quires that  certain  changes  be  made  to  the  basic 
image  forming  equations  for  both  reflection  and 
transmission  modes  of  imaging.    We  will  show  that 
consideration  of  these  requirements  provides  not 
only  a  method  for  imaging  structures  embedded  in 
the  moving  media  (points  in  the  moving  media)  but 
also  methods  for  imaging  the  velocity  distribu- 
tion of  the  moving  media  itself. 

The  equations  of  synthetic  focusing  can  be 
adapted  to  the  case  of  moving  media  by  using  the 
time  of  travel  associated  with  the  ray  paths 
which  have  been  bent  by  the  moving  media.  Thus, 
the  expression  for  the  time  t  of  travel  must  be 
replaced  by  an  expression  containing  the  velocity 
of  the  media.    For  the  case  where  the  velocity  of 
the  media  is  small  compared  to  the  acoustic  speed, 
this  may  be  written  as  (by  modifying  equation  3) 

m  k 

Here,  u  is  acoustic  speed,  \f  is  fluid  velocity, 
and  T  is  a  unit  tanget  vector  along  the  ray  path 
connecting  k  to  the  transducer  elements  m  or  j.  A 
more  exact  expression  for  t,  in  the  case  when  t 
is  not  very  small  in  comparison  to  u,  has  been 
given  by  Johnson  and  others  [10,13].    In  addition 
to  time  shifts,  a  moving  media  also  produces  a  com- 
pression or  expansion  of  the  shape  of  a  waveform. 
When  more  exact  imaging  is  required,  the  effect  of 
this  stretching  may  be  removed  by  use  of  an  appro- 
priate compensating  algorithm  as  is  now  shown. 
Let  g(y)  =  b  •  h(t-to)  be  the  received  synthetic 
focus  waveform  for  \?  =  0.    Here,  b  is  the  scat- 
tering strength  and  h  is  the  transmitted  waveform 
(for  simplicity,  the  frequency  dependence  of  the 
scattering  process  is  assumed  to  be  found  in 
H  (•))•    Let  an  upper  case  letter  represent  the 
Fourier  transform  of  a  corresponding  lower  case 
letter.    Then,  the  corresponding  expression  in 
frequency  space  is  G(f)  =  b  •  exp(-2iTrfto)H(f ) . 

The  Doppler  frequency  shift  due  to  motion  of 
the  media  has  the  effect  of  replacing  f  by  (f  +  Af). 
I^lere,  Af  is  given  by  Af  =  f  ^  •  (Tr  +  Ts)/u,  where 
V  is  the  velocity  of  the  flow  and  where  Tr  and  Ts 
are  the  unit  tanget  vectors  from  the  scattering 
center  in  the  flow  to  the  receiver  and  transmitter 
(source)  respectively.    For  |V|/u  <<1,  the  effect 
of  this  frequency  shift  changes  the  form  of  G(f) 


to  GD(f)  given  by  GD(f)  =  b  •  exp(-2^fto)H(l  -  6)(f) 
where  6f  =  Af.    The  corresponding  time  functions 
are  given  by 

gp(t)  =  (b/|l  -  6|)h((t  -  to)/(l  -  6)).  (6) 

Thus,  movement  of  a  scattering  center  toward  a 
source  and  receiver  produces  a  shortening  of  pulse 
width  and  an  increase  in  pulse  amplitude. 

B.    Synthetic  Focus  Doppler  Imaging 

The  principle  of  synthetic  focusing  may  be  ap- 
plied to  Doppler  velocity  imaging  to  improve  the 
spatial  and  velocity  resolution  and  to  capture 
more  of  the  Doppler  scattered  signal  (by  the  use 
of  a  larger  receiving  aperture)  and  thereby  in- 
creasing the  signal-to-noise  ratio  in  the  final 
velocity  image.    Simply  increasing  the  aperture 
of  a  single  transducer  may  not  improve  Doppler 
resolution  and  signal-to-noise  ratio  because  the 
Doppler  shift  will  not  be  constant  over  the  large 
aperture.    A  method  is  needed  which  will  make 
use  of  the  different  Doppler  shifts  on  a  large 
sampled  aperture. 

For  simplicity,  a  method  suitable  for  flows  re- 
stricted to  a  plane  will  be  given  first.    It  is 
assumed  that  the  transmitting  and  receiving  trans- 
ducers are  also  restricted  to  this  plane  and  are 
located  on  part  of,  or  on  the  complete,  circum- 
ference of  a  circle.    Assume  that  the  transmitter 
produces  a  narrow  beam  of  pulsed-energy  which 
constitutes  a  cord  of  this  circle.    Let  the  velo- 
city of  the  flow  at  e|ch  point  along  the  cord  be 
^(q).    Let  Ts(q)  and  Tj(q)  be  the  unit  tanget 
vectors  from  the  scattering  center  q  to  the  trans- 
mitter source  s  and  receiver  j  respectively.  The 
use  of  time  gates  to  achieve  spatial  resolution  is 
assumed.    Then  the  Doppler  shift  Af  is  given  by 
(f/u)V  •  (Ts  +  Tj).    This  may  be  written  as 
Af(q)  =  (f/u(q))  [(cos  ot  +  cos  ej)Vx  +  (sin  et  + 
sin  ej)Vy]q.    Then  this  constitutes  a  set  of 
simultaneous  equations  in  Vx  and  Vy.    This  set  is 
usually  over  determined  since  the  set  can  be 
spanned  with  only  two  independent  values  of  e j , 
but  the  over-determination  provides  for  better 
signal-to-noise  level.    The  above  equation  may 
also  be  written  as 

Af(q)j  =  [f/u(q)]  (|V(q)|)  [cos(\r,fs)  (7) 

+  COs(V,Tj)],  j  =  1,2... 

Note  that  the  first  term  is  a  frequency  bias 
and  the  second  term  has  a  period  of  2-n  in  ej. 
The  direction  of  the  flow  ?  and  its  magnitude  may 
be  found  from  this  set  of  equation  by  several 
schemes : 

1)  The  Fourier  transform  of  the  |.bQve  equation 
may  be  taken  with  respect  to  angle  (V,Tj).  The 
square  root  of  the  sum  of  the  squares  of  sin  and 
cos  terms  with  pet;ii}d  2-n  is  equal  to  (|V|f/u) 
while  the  angle  (V,Ts)  is  proportional  to  the  arc 
tangent  of  the  ratio  of  these  terms. 

2)  A  least  squares  fit  of  a  phase  shifted 
cosine  function  can  be  made  to  the  data.  This 
least  squares  technique  can  also  be  applied  to 
three-dimensional  flows  and  two-dimensional  de- 
tectors.   The  velocity  V  can  then  be  obtained. 

These  two  schemes  and  the  theory  presented  up 
to  this  point  assume  that  a  unique  value  of  Af 
may  be  obtained  for  each  point  q  and  detector  j. 


242 


This  is  not  exactly  true  because  in  practice,  the 
frequency  measure  is  broadened  due  to:    1)  noise, 
2)  ambiguity  due  to  range  gating,  and  3)  turbu- 
lence in  the  flow.    In  most  or  many  cases,  the 
broadening  may  be  tolerated  by  the  use  of  offset 
frequency  quadrature  Doppler  detection  schemes 
which  produce  an  output  voltage  whose  mean  is 
proportional  to  the  mean  Doppler  shift  of  a 
broadened  signal  [3]. 

C.    Flow  Reconstruction  by  Acoustic  Transmission 

We  have  previously  suggested  that  fluid  flow 
within  a  measurement  region  may  be  determined  by 
transmitting  and  receiving  acoustic  energy  through 
the  measurement  region  along  a  plurality  of  rays 
such  that  each  volume  element  is  transversed  by  a 
set  of  rays  having  components  in  each  direction 
for  which  flow  components  are  to  be  reconstructed 
[10].    The  propagation  time  of  the  acoustic  energy 
along  the  plurality  of  rays  constitutes  the  only 
measurements  required  by  this  method  which  has  now 
been  verified  for  flows  with  velocities  small  com- 
pared to  the  speed  of  sound  [4].    Since  this 
method  is  analogous  to  computed  reconstruction 
tomography  methods  and  has  been  reported  elsewhere, 
the  corresponding  theory  is  not  reported  here. 

D.    Synergistic  Flow  Reconstruction 

It  is  clear  that  both  transmission  reconstruc- 
tion and  Doppler  methods  could  be  applied  simul- 
taneously with  sufficient  parallel  data  collection 
and  processing  capability.    Such  an  approach,  com- 
bining larger  apertures  and  both  imaging  modes, 
although  perhaps  impractical  at  this  time,  has 
several  theoretical  advantages:    1)  greater  signal- 
to-noise  ratios,  2)  greater  spatial  and  velocity 
resolutions,  3)  the  transmission  reconstruction 
mode  can  provide  refractive  index  information  for 
correction  of  the  Doppler  imaging  mode,  and  4) 
drag  velocity  might  be  calculated  from  the  dif- 
ference between  the  Doppler  mode  velocity  image 
(scattering  center  velocity)  transmission  mode 
velocity  image.    These  principles  could  be  applied 
to  statistically  steady-state  flow  using  one 
Doppler  channel  and  one  transmission  channel  by 
time  multiplex  methods. 

5.    Computer  Simulation  and 
Experimental  Studies 

An  example  of  the  resolution  capability  of  the 
synthetic  focus  technique  is  demonstrated  in 
[    figure  7.    This  figure  is  very  significant  because 
I    it  presents  experimental  evidence  that  resolution 
1    of  less  than  one-half  wavelength  (at  the  center 
I    frequency  of  a  pulse)  can  be  obtained  with  ultra- 
1    sound;  at  3  MHz  this  corresponds  to  0.25  rrm.  It 
I    is  expected  that,  when  synthetic  focusing,  refrac- 
I    tive  index  reconstruction  and  other  ultrasound 
imaging  modalities  are  combined,  a  synergistic 
union  will  result.    Thus,  each  mode  will  not  only 
add  its  own  form  of  new  information  but  will  also 
help  remove  some  of  the  limitations  of  other 
'  forms  of  imaging.    For  example,  knowledge  of  re- 
)   fractive  index  can  correct  for  defocusing  effects 
i    in  echo  imaging  mode  [9,11]. 

Thus,  a  major  limitation  to  applying  the  com- 
1    plete  capabilities  of  the  synthetic  focus  tech- 
nique to  complex  tissues  can  be  solved  by  the 
synergistic  treatment  of  the  problem  of  refrac- 


raw  data  deconvolved  data 


Fig.  7    Synthetic  focus  images  of  a  nylon  thread 
from  data  taken  from  one  view  angle  and 
from  multiple  view  angles.    The  top  left 
image  represents  the  raw  data  from  26  re- 
ceiver elements  for  each  of  five  trans- 
mitters.   Left  image  A  is  the  correspond- 
ing synthesized  image.    The  lower  left 
image  B  is  obtained  from  simulating  the 
effect  of  many  such  32  element  arrays  on 
a  circle  or  radius  corresponding  to  geo- 
metry of  image  A.    Top  right  is  the  re- 
sult of  deconvolving  the  ray  data  of  top 
left  with  impulse  response  of  the  system. 
Middle  right  A  and  lower  right  are  images 
synthesized  from  deconvolved  data  and  cor- 
responding to  A  and  B,  respectively.  The 
labeled  line  segments  and  circles  in  the 
center  are  the  relative  wavelength  of 
sound  drawn  to  the  same  scale  as  the  re- 
constructed images.    Note  that  the  full 
width  at  half  maximum  of  the  peak  in  image 
B'  (FWHMg')  is  less  than  one-half  wave- 
length (a/2)  while  that  in  image  B  is 
larger  and  more  nearly  A/2.    Each  picture 
element  is  .04  mm  square.    The  resonant 
frequency  of  the  transducer  elements  is 
about  3  MHz.     (Reproduced  with  permission 
from  Johnson  et  al . ,  Digital  Processing 
of  Biomedical  Images,  1976,  pp.  203-226.) 

tion  of  ultrasound.    Consequently,  an  iterative 
correction  of  refraction  has  been  developed.  This 
ray  tracing  technique  has  also  been  approached 
from  established  geophysical  seismic  imaging 
theory.    The  corrections  used  by  seismologists  to 
unscramble  their  seismographic  results  indicate 
the  feasibility  of  this  approach.    An  example  of 
the  order  of  improvement  attending  images  origi- 
nally collected  by  synthetic  focus  methods,  when 
seismic  image  correction  techniques  are  used,  is 
given  in  figure  8.    This  image  is  significant  be- 
cause it  represents  (to  our  knowledge)  the  first 
application  of  seismic  wave  migration  techniques 
to  real  ultrasound  data.    This  image  has  been 
presented  in  an  earlier  paper  [9]. 

In  these  images  presented  in  figures  7  and  8, 
no  attempt  has  been  made  to  accurately  control 
the  waveform  of  the  echo  from  a  single  point  to 
maximize  resolution.    Although,  the  right-hand 


243 


plastic  wedge  part  of  "MAYO" 


Fig.  8    Refraction  corrected  synthetic  focus  image. 
The  bottom  portion  of  figure  shows  a 
plastic  wedge  and  pattern  of  nylon  threads 
which  spell  "Mayo",  however,  only  the  "MA" 
portion  is  visible  in  the  figure.    An  ar- 
ray of  32  transducer  elements  is  located 
to  the  left  but  is  not  shown.    Sonic  waves 
produced  by  the  array  refract  through  the 
wedge,  reflect  or  scatter  from  the  threads, 
pass  back  through  the  wedge  and  are  then 
received  by  the  array.    No  recognizable 
image  was  produced  when  the  corresponding 
data  was  processed  with  a  simple  synthetic 
focus  algorithm  which  assumed  constant 
speed  of  the  ultrasound  waves.    The  out-of- 
focus  image  produced  by  this  simple  algo- 
rithm is  not  shown  here. 

The  image  at  the  top  is  the  result  of  ap- 
plication of  a  geophysical  data  processing 
algorithm  which  corrects  for  refraction 
without  prior  information  of  the  presence 
of  the  wedge.    This  is  accomplished  by 
"wave  front  migration"  techniques  which 
determine  the  effective  refractive  index 
layer  by  layer  as  the  algorithm  works  it- 
self away  from  the  array.    Such  algorithms 
work  for  layered  materials  but  could  be  ex- 
tended to  structures  found  in  the  human 
body.    The  center  frequency  of  the  damped 
burst  transmitted  from  the  array  is  3  MHz. 
The  target  is  the  same  one  reported  in 
Volume  6  of  Acoustical  Holography  by  S.  A. 
Johnson  et  aT.  [7 ] . 

side  of  figure  7  shows  the  image  produced  by  a 
deconvolution  process  which  satisfies  only  the 
general  requirements.    Therefore,  a  computer 
simulation  study  was  undertaken  to  investigate 
the  maximum  2-D  resolution  limit  which  can  be 
obtained  with  a  signal  constrained  to  have  no 
frequency  components  greater  than  an  upper  fre- 
quency limit  fm-    This  transmitted  kernel  signal 
h(t)  was  chosen  to  have  a  frequency  distribution 


given  by  H(f)  =  Hglfl  for  |f|  <  fm  and  H(f)  = 
0  for  |f|  >  fm.    This  function  is  the  well-known 
Ramachandran  Lakshminaraynan  kernel  [iSl.    A  data 
collection  geometry  of  a  12  cm  radius  ring  of  120 
common  transmitter  and  receiver  positions  equally 
spaced  on  360  degrees  was  simulated.  Imaging 
was  confined  within  a  concentric  7  cm  circle.  The 
value  of  fjTi  was  set  at  1  MHz  and  the  simulated  re- 
ceived echo  signals  were  sampled  at  100  ns  inter- 
vals.   The  simulated  received  echo  signals  were 
back  projected  along  ellipses  in  the  image  space 
as  per  figure  4.    The  2-D  point  response  function 
(2-D  PRF)  was  found  not  to  differ  significantly 
between  points  near  the  center  and  edge  of  the 
7  cm  radius  imaging  region.    Various  (time)  fre- 
quency filtered  versions  of  the  above  kernel  were 
tried  also.    Only  slight  improvement  in  2-D  PRF 
was  observed  with  some  kernels  but  reduction  in 
2-D  side  lobe  response  could  be  obtained  with  an 
attendant  slight  loss  in  resolution  with  other 
kernels.    Some  of  the  results  of  these  studies 
are  presented  in  figure  9. 

Since  the  experimental  image  obtained  by  ap- 
plication of  the  new  transmission  mode  fluid 
temperature  and  vector  velocity  technique  have 
been  published  in  more  detail  previously  [4],  only 
one  example  reconstruction  is  presented  here. 
Figure  10  shows  reconstructions  of  fluid  flows 
made  in  the  refractive  index  reconstruction  breast 
scanner  at  Mayo  Clinic.    Applications  of  these 
techniques,  for  example,  for  measurement  and  re- 
construction of  tissue  temperatures  during  hyper- 
thermia cancer  therapy  have  been  reported  else- 
where [16]. 


Data  on  X.i£.2  Image  on 


Fig.  9    Illustration  of  simulated  data  and  cor- 
responding synthetic  focus  resolution  test. 
Top  left:    gray  scale  plot  of  received 
data  for  two  scattering  centers  separated 
by  1/2  wavelength  (measured  at  the  maxi- 
mum frequency  limit  of  transmitted  kernel). 
The  amplitude  for  120  views  on  360°  vs 
time  is  shown.    Bottom  left:  amplitude 
plot  of  above  data  along  brightened  line 
l]^l2.    Top  right  -  0  mm  by  9.6  mm  synthetic 
focus  image  (64  by  64  pixels)  from  these 
data.    Bottom  right:    amplitude  plot  along 
the  white  line  I3I4  in  the  above  image 
demonstrating  the  resolution  of  the  two 
point  targets.    In  the  gray  scale  images 
above,  gray  =  zero,  black  =  negative,  and 
white  =  positive  amplitudes. 


244 


Fig.  10.    Experimental  data  and  reconstruction  of  vector  components  of  fluid 
vortex.    Top  left  shows  experimental  data  and  is  an  image  of  the 
difference  between  the  time  of  arrival  with  flow  and  without  flow 
(fast  arrival  =  white,  no  change  gray,  slow  =  black)  vs^.  scan 
position  (left  to  right)  vs_.  angle  of  view  (top  to  bottom).  Top 
right  shows  reconstructed  planar  fluid  speed  (v^  +  V^)''5,  black  is 
zero,  white  is  positive.    Bottom  left  shows  x  component  of  velocity 
Vx-    Bottom  right  shows  y  component  of  velocity  Vy.    In  Vx  and  Vy, 
black  is  negative,  gray  is  zero  and  white  is  positive.  Reconstruced 
flow  is  maximum  (73  cm/s)  at  a  radium  of  0.62  cm.  Reconstructions 
are  64  pixels  per  side.    Geometry  of  vortex  scanning  plane,  and 
vortex  generator  are  shown  in  top  and  side  views  in  right  margin. 
(Reproduced  with  permission  from  Greenleaf  et  al . ,  Quantitative 
Imaging  from  Transmission  Ultrasound  (in  press ) . ) 


6.  Conclusion 

We  suggest  that  instruments  capable  of  produc- 
ing higher  resolution  images  of  specific  ultra- 
sound tissue  parameters  would  provide  a  valuable 
service  in  clinical  applications.    Wide  aperture 
techniques  are  known  which  can  provide  this  high 
resolution.    To  be  successfully  applied  in  the 
human  body,  these  wide  aperture  techniques  must 
be  modified  to  compensate  for  refraction  and  pos- 
sibly, in  some  cases,  also  diffraction  effects. 
Candidate  techniques  can  be  found  in  the  synthetic 
focusing  methods  developed  in  our  laboratory  and 
in  analogous  geophysical  or  seismic  methods. 
These  techniques  can  provide  resolution  of  one- 
half  wavelength  in  homogeneous,  isotropic  media. 
It  seems  reasonable  that  this  degree  of  resolu- 
tion may  De  approached  in  the  body  by  use  of  wide 
apertures  and  improved  refraction  and  absorption 
compensating  algorithms.    Improvement  of  resolu- 
tion from  one  wavelength  to  one-half  wavelength 
is  difficult  because  knowledge  of  the  spatial  and 
temporal  history  of  the  transmitted  and  received 
waveform  is  necessary  for  application  of  these 
algorithms.    This  knowledge  may,  in  principle, 
be  obtained  with  the  proper  transducer  design  and 
scanner  geometry.    Our  computer  simulation  and 
laboratory  experiments  provide  encouragement  and 
evidence  that  resolutions  of  about  one  to  one- 
half  wavelength  should  be  feasible  in  some  ap- 
plications.   We  further  predict  that  efforts  to 
design  economical  scanners  to  provide  this  resolu- 
tion will  be  a  fruitful  research  and  development 


endeavor.    We  believe  such  endeavors  will  rely 
heavily  upon  the  use  of  high  speed  digital  data 
processing  methods  and  sophisticated  mechanically 
and  electronically  scanned  apertures  with  large 
solid  angles. 

Acknowl edgment 

The  support  from  Dr.  Earl  H.  Wood,  Dr.  Erik 
Ritman,  and  the  staff  of  the  Biodynamics  Research 
Unit,  Mayo  Clinic,  is  appreciated.    The  refraction 
corrected  image  of  the  "MA"  shown  in  figure  9  was 
computed  using  seismic  migration  techniques  with 
the  help  of  John  Parr,  of  Houston,  Texas,  and  his 
help  is  greatly  appreciated.    This  research  was 
supported  by  Grants  HL-00170,  HL-00060,  RR-000Q7. 
Hl-04664,  NIH-HT-42904  from  the  National  Insti- 
tutes of  Health,  United  States  Public  Health 
Service;  NCI-CB-64041  from  the  National  Cancer 
Institute. 


References 

[1]    Wolfe,  J.  N. ,  Risk  for  breast  cancer  develop- 
ment determined  by  mammographic  parenchymal 
pattern.  Cancer  37.  2486-2492  (1976). 

[2]    Wolfe,  J.  N.,  Breast  patterns  as  an  index  of 
risk  for  developing  breast  cancer.  Am.  J. 
Roentgenol.  126,  1130-1139  (1976). 


245 


[3]    Woodcock,  J.  P.,  Development  of  the  ultra-  [lO] 
sonic  flowmeter.  Ultrasound  in  Medicine  and 
Biology  2.  11-18  (1975). 

[4]    Johnson,  S.  A.,  Greenleaf,  J.  F. ,  Tanaka,  M., 
and  Flandro,  G.,  Reconstructing  three- 
dimensional  temperature  and  fluid  velocity 
vector  fields  from  acoustic  transmission  [ll] 
measurements.    Instrument  Society  of  America 
Transactions,  March  1977  (in  press). 

[5]    Sandler,  H.  and  Rasmussen,  D. ,  Angiographic 

analysis  of  heart  geometry,  in  Roentgen- ,  [i2] 
Cine-,  and  Videodensitometry,  Paul  H. 
Heintzen,  ed.,  pp.  212-223  (Georg  Thieme 
Verlag,  Stuttgart,  1971). 

[6]    Cardiovascular  applications  of  ultrasound. 
Proceedings  of  an  International  Symposium 
held  at  Janssen  Pharmaceutics,  Beerse, 
Belgium,  May  29-30,  1973.    Robert  S.  Reneman, 
ed.  (American  Elsevier  Publishing  Company, 
Inc.,  New  York,  1974).  [13] 

[7]    Johnson,  S.  A.,  Greenleaf,  J.  P.,  Duck, 

F.  A.,  Chu,  A.,  Samayoa,  W.  F.,  and  Gilbert, 

B.  K.,  Digital  Computer  Simulation  Study  of  [14] 

a  Real-Time  Collection,  Postprocessing 

Synthetic  Focusing  Ultrasound  Cardiac  Camera, 

in  Acoustical  Holography,  Newell  Booth,  ed.. 

Vol.  6,  pp.  193-211  (Plenum  Press,  New  York, 

1975). 

[15] 

[8]    Duck,  F.,  Johnson,  S.  A.,  Greenleaf,  J.  F. , 
and  Samayoa,  W. ,  Digital  image  focusing  in 
the  near  field  of  a  sampled  acoustic  aper- 
ture. Ultrasonics,  83-88  (March,  1977). 

[9]    Johnson,  S.  A.,  Greenleaf,  J.  F.,  Bahn, 

R.  C,  Samayoa,  W.  F.,  Duck,  F.,  Hansen,  C,  [16] 
Tanaka,  M.,  and  Lent,  A.,  Reconstruction  of 
three-dimensional  ultrasound  tissue  proper- 
ties from  transmission  and  other  data. 
American  College  of  Radiology  Conferesce  on 
Computerized  Tomography  and  Radiology,  St. 
Louis,  Missouri,  April  25-28,  1976,  pp. 
149-162. 


Johnson,  S.  A.,  Greenleaf,  J.  F.,  Samayoa, 
W.  F.,  Duck,  F.  A.,  and  Sjostrand,  J., 
Reconstruction  of  three-dimensional  velocity 
fields  and  other  parameters  by  acoustic  ray 
tracing.    1975  Ultrasonics  Symposium  Proceed- 
ings, IEEE  Cat.  #75  CHO  994-4SU. 

Claerbout,  J.  F.,  Fundamentals  of  Geophysi- 
cal Data  Processing  with  Applications  to 
Petroleum  Prospecting  (McGraw-Hill,  New 
York,  1976). 

Greenleaf,  J.  F.,  and  Johnson,  S.  A., 
Algebraic  Reconstruction  of  Spatial  Distri- 
butions of  Refractive  Index  and  Attenuation 
in  Tissues  from  Time-of-Fl ight  and  Amplitude 
Profiles,  in  Ultrasonic  Tissue  Characteriza- 
tion, M.  Linzer,  ed.,  National  Bureau  of 
Standards  Spec.  Publ.  453,  pp.  109-119, 
October,  1976.     (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

White,  R.  W.,  Acoustic  ray  tracing  in  moving 
inhomogeneous  fluids.  Acoustical  Society  of 
America  53  (6),  1700-1703  (1973). 

Brooks,  Rodney  A.  and  Di  Chiro,  Giovanni, 
Principles  of  computer  assisted  tomography 
(CAT)  in  radiographic  and  radioisotopic 
imaging,  Phys.  Med.  Biol.  2j[  (5),  689-732 
(1976). 

Ramachandran ,  G.  N.,  and  Lakshminarayanan , 

A.  v..  Three-dimensional  reconstruction  from 
radiographs  and  electron  micrographs:  ap- 
plication of  convolutions  instead  of  Fourier 
transforms,  Proc.  Natl.  Acad.  Sci.,  U.S.  68 
(9),  2236-2240  (1971). 

Johnson,  S.  A.,  Greenleaf,  J.  F.,  Rajago- 
palan,  B.,  Christensen,  D.  A.,  and  Baxter, 

B.  ,  Nonintrusive  acoustic  temperature  tomo- 
graphy for  measurement  of  microwave  and 
ultrasound-induced  hyperthermia.  Workshop 
on  Electromagnetics  and  Cancer,  1977  lEEE- 
MTTS  International  Microwave  Symposium, 
June  24,  1977.     (Submitted  to  the  Journal 
of  Bioengineering . ) 


246 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


MAPPING  TRUE  ULTRASONIC  BACKSCATTER  AND  ATTENUATION  DISTRIBUTION  IN  TISSUE  - 
A  DIGITAL  RECONSTRUCTION  APPROACH 


F.  A.  Duck  and  C.  R.  Hill 

Physics  Divisiorij  Institute  of  Cancer  Research 
Clifton  Avenue,  Sutton,  Surrey,  United  Kingdom 


The  back-scattered  intensity  from  any  location  in  an  ultrasonically  irradiated 
medium  is  dependent  upon  both  the  local  back-scattering  cross-section  and  the  attenua- 
tion by  the  overlying  medium.    An  iterative  digital  reconstruction  technique  has  been 
investigated  which  is  capable  of  mapping  quantitatively  and  separately  the  distribu- 
tions of  attenuation  coefficient  and  back-scatter  cross-section  using  multiple  pulse- 
echo  data.    Such  a  technique  will  enable  tissue  to  be  characterised  on  the  basis  of 
these  two  separate  acoustic  parameters.    A  description  of  the  technique  is  given, 
together  with  results  from  simulation  studies  leading  to  an  improved  processing  tech- 
nique.   The  potential  and  limitations  of  the  method  are  discussed. 


Key  words:    Attenuation;  back-scattering  cross-section;  digital  reconstruction; 
iterative;  ultrasound. 


1.  Introduction 

A  major  deficiency  of  currently  used  pulse-echo 
techniques  for  tissue  characterisation  and  imaging 
is  their  failure  to  give  truly  quantitative  mea- 
surements of  tissue-specific  acoustic  parameters. 
The  problem  arises  since  the  observed  time- 
dependent  echo  amplitudes  result  from  two  sets  of 
unknown  quantities.    These  are  back-scattering 
cross-sections  of  the  interrogated  volumes  and 
attenuation  coefficients  of  the  volume  in  the 
transmission  path.    In  conventional  B-Scan  prac- 
tice it  is  usual  to  emphasise  the  use  of  back- 
scattering  cross-section  as  the  useful  tissue 
characterisation  parameter,  and  suppress,  or  com- 
pensate for,  the  amplitude  changes  due  to  the  tis- 
sue attenuation.    This  latter  is  done  in  two  ways. 
In  the  first  place,  a  time-gain-compensation  (TGC) 
amplifier  is  used  which  compensates  for  attenuation 
under  the  assumption  that  attenuation,  if  not  uni- 
form, varies  in  a  way  which  is  identical  irrespec- 
tive of  transducer  position  and  orientation.  The 
precise  form  of  the  attenuation  versus  time  (or 
depth)  variation  is  often  under  operator  control. 
Secondly,  compound  scanning  techniques  are  used 
which  result  in  an  evening-out  of  the  shadowing 
effects  of  overlying  tissue  layers,  with  any  given 
TGC  function.    Automatic  methods  have  been  de- 
veloped for  adaptive  gain  compensation  (McDicken 
et  al.  [l]i;  De  Clercq  et  al.  [2]).    These  have 
shown  that  some  improvement  in  gain  compensation 
can  be  obtained,  and  presumably  the  TGC  function 
generated  could  be  used  diagnostical ly.  However 
such  techniques  are  designed  to  operate  on  multiple 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


pulse-echo  data  taken  from  a  particular  orientation, 
assume  uniform  back-scattering  and  have  an  operat- 
ing time  constant  of  several  seconds.    In  contrast, 
the  method  we  have  developed,  and  which  is  describ- 
ed below,  generates  separately  two-dimensional 
maps  of  attenuation  coefficient  and  back-scatter- 
ing cross-section,  from  multiple  orientated  pulse- 
echo  data  which  may  be  gathered  very  rapidly  for 
subsequent  processing.    The  two  clear  incentives 
for  investigating  this  technique  are: 

1)  It  provides  a  quantitative  map  of  acoustic 
attenuation  coefficients  which  may  be  used  dfag- 
nostically  for  tissue  characterisation. 

2)  It  provides  a  correctly  attenuation- 
compensated  B-Scan,  a  map  which  can  now  be  correct- 
ly referred  to  as  a  back-scatter  map. 

2.    The  Method 

As  noted  above  little  or  no  diagnostic  informa- 
tion from  local  attenuation  variations  is  avail- 
able when  compound  B-Scanning  is  used.    In  simple 
scanning,  however,  regions  which  differ  signifi- 
cantly from  the  expected  local  attenuation  will, 
by  virtue  of  the  pre-set  TGC  used  throughout  the 
scan,  result  in  modified  signal  levels  from  re- 
gions beyond  them.    For  example  a  local,  dense 
athero-sclerotic  lesion  will  cast  a  shadow  below 
it,  and  a  fluid  filled  cyst  of  low  attenuation  will 
result  in  increased  amplitude  signals  from  tissue 
beyond  it  since  the  gain  is  over-compensating  for 
attenuation  losses.    Such  signs  are  used  clinically 
for  diagnosis  but  only  under  those  circumstances 
where  the  local  effects  are  clear.    The  basic  con- 
cepts are  shown  simply  in  diagrammatic  form  in 
figure  1,  which  shows  images  of  a  localised  in- 
crease in  attenuation  (at  A)  in  a  uniform  scatter- 
ing medium.    A  simple  linear  scan  results  in  an 


247 


Primary  images 


a)  Linear  scan  1  b)  Linear  scan  2 

Secondary  images 


c)  Scatter  map  d)  Attenuation  map 


vestigated  in  ultrasound  for  the  reconstruction  of 
acoustic  attenuation  from  transmitted  pulse  ampli- 
tudes (Greenleaf  et  al .  [3])  and  for  the  reconstruc- 
tion of  acoustic  refractive  indices  from  transmit- 
ted pulse  times-of-fl ight  (Greenleaf  et  al .  [4]). 
The  analytic  description  for  this  problem  was 
formulated  by  Radon  [5]  and  a  range  of  computation- 
al techniques,  based  on  this  analysis  or  on  the 
use  of  iterative  techniques,  have  been  reported 
(e.g. ,  Gordon  et  al .  [6]).    In  transmission  imaging 
the  reduction  in  intensity  (or  shadow)  in  the 
transmitted  energy  can  be  measured  directly.  In 
the  present  situation  however  this  reduced  intensi- 
ty signal  cannot  be  measured  directly  but  is  re- 
radiated  with  a  strength  depending  upon  the  local 
back-scatter  cross-section.    A  new  set  of  unknown 
quantities  is  therefore  introduced.    Although  the 
values  of  these  back-scattered  cross-sections  are 
unknown  their  related  positions  in  space  are  known 
since  the  position  and  orientation  of  the  trans- 
ducer and  time  delay  can  be  measured. 

3.  Theory 


Fig.  1.    The  generation  of  separate  maps  of  back- 
scatter  cross-section  and  attenuation  co- 
efficient from  multiple  pulse  echo  data. 
A  local  increase  in  attenuation  at  A  with- 
in a  scattering  medium  casts  shadows  S  and 
T  in  simple  linear  scan  images  (a)  and  (b). 
The  combination  of  these  by  peak  detection 
gives  a  first  guess  at  a  scatter  map  (c). 
The  differences  as  between  this  and  a,  b 
are  used  to  generate  d,  the  attenuation 
coefficient  map  and  correct  c. 

image  (fig.  la)  in  which  a  shadow  (S)  is  cast 
through  the  image  behind  the  local  attenuating 
site.    In  this  simple  case  it  would  normally  be 
assumed  that  there  was  indeed  a  locally  high  at- 
tenuation at  A.    However,  an  identical  image  could 
have  been  generated  if  there  was  an  elongated  re- 
gion of  low  back-scatter  cross-section  which  was 
coincident  with  the  shadow  region  S.    These  two 
situations  can  be  easily  distinguished  by  carry- 
ing out  a  second  simple  linear  scan  at  a  new  angle 
(9)  (fig.  lb).    Now  the  region  which  was  in  shadow 
(S)  is  imaged  clearly  and  there  is  a  new  shadow 
(T).    Combining  the  two  simple  scans  gives  a  com- 
pounds scan  (fig.  Ic)  which  may  destroy  the  at- 
tenuation information.    Comparing  the  simple  and 
compound  scans  enables  the  true  attenuation  map  to 
be  inferred. 

In  general  of  course,  all  tissue  layers  will 
cast  an  acoustic  shadow  on  tissue  layers  behind 
them  to  a  greater  or  lesser  extent,  and  the  real 
situation  is  additionally  complicated  because  the 
scattering  characteristics  of  the  medium  vary. 
However,  as  seen  below,  account  can  be  taken  of 
both  sets  of  parameters  simultaneously. 

The  physical  situation  described  is  in  many 
ways  conceptually  similar  to  the  problem  handled 
in  axial  tomography,  and  whose  computed  solutions 
are  exploited  widely  in  clinical  medicine  for  the 
reconstruction  of  x-ray  attenuation  coefficients 
from  transmitted  x-ray  intensities.    Here  the 
problem  of  the  reconstruction  in  3  dimensions  of 
a  parameter  map  from  a  set  of  2-dimensional  pro- 
files is  reduced  to  that  of  the  reconstruction  of 
a  set  of  parallel,  2-dimensional  distributions  of 
the  parameter  from  related  1-dimensional  profiles. 
Such  transmission  imaging  techniques  have  been  in- 


The  mathematical  formulation  of  the  problem  is 
designed  to  establish  a  set  of  simultaneous  equa- 
tions based  upon  an  assumed  model;  the  equations 
may  then  be  solved  iteratively  by  computational 
techniques.    At  present  there  is  no  equivalent 
formulation  to  that  of  Radon  for  the  reconstruc- 
tion of  distributions  of  parameters  from  transmit- 
ted profiles.    Indeed  it  may  well  be  that  if  there 
were  it  would  not  be  of  direct  use  computationally 
because  of  the  necessary  assumptions  made  in  the 
model . 

The  back-scattered  intensity  Ip  received  from 
any  position  r  in  an  ultrasonically  irradiated 
field,  can  be  given  as 


I    =  I 


ref 


A(r) 


exp 


a  It' 
r 


(1) 


where  A(r)  is  a  factor  depending  upon  the  beam 
geometry; 

r  =  the  distance  of  the  scattering  target  from 

the  acoustic  centre  of  the  transducer; 
c  =  the  local  attenuation  coefficient;  and 
a  =  the  local  back-scattering  cross-section  per 
unit  volume  per  steradian. 
In  logarithmic  forms  this  becomes 


2.n 


(2) 


In  order  that  the  situation  is  amenable  to  a  digi- 
tal solution  the  region  to  be  imaged  can  be  model- 
led as  a  rectangular  grid  of  n  elements,  within 
each  of  which     and  a    are  constant,  with  the  only 
changes  being  at  the  boundaries  of  the  elements. 
This  geometry  is  illustrated  in  figure  2.    In  ad- 
dition it  is  assumed  that  there  are  no  losses  from 
absorption  and  scattering  processes  outside  the 
region  to  be  imaged.    Then,  for  any  image  element 
within  this  set,  eq.  (2)  can  be  rewritten  as 


In 


(Vef) 


=  K 


2L  +  ^n(0^ 

n=l 


(3) 


248 


— 

— i 

i 

Transducer 


Fig.  2.    Image  grid  geometry.    The  image  consists 
of  n  cells  Pi...Pn.    Each  cell  Pf  has  as- 
sociated with  it  acoustic  parameters  of 
attenuation  coefficient  a-j  and  back- 
scatter  cross-section,  o-jWp^  is  the  ray/ 
cell  intersection  length. 

where  K,-  is  a  potentially  known  geometric  factor, 
and  where  W-j ,  i  =  l...n,  are  a  set  of  geometric 
weighting  factors  related  to  the  image  grid  and 
beam  geometry.    If  ray  geometry  is  assumed  then 
W-j  is  the  length  of  intersection  of  the  ray  with 
the  image  element  Pi.    In  general  W-j  =  0  for 
many  i. 

If  a  multiplicity  of  measurements  1^  are  made, 
from  a  number  of  directions  then  eq.  (3)  extends 
to  a  set  of  simultaneous  equations  in  which  a-j, 
i  =  l...n  and  a-j,  i  =  l...n  are  the  unknowns. 
One  characteristic  of  this  set  of  equations  is 
that  it  is  sparse  since,  as  pointed  out  above, 
many  values  of  Wi  are  zero.    A  digital  iterative 
technique  is  suited  to  the  solution  since  it  can 
handle  such  large  sets  of  equations  under  condi- 
tions where  exact  solutions  may  be  multiple,  or 
not  exist  at  all.    As  pointed  out  above,  such 
techniques  have  already  been  used  highly  success- 
fully in  medicine  for  the  reconstruction  of  the 
distribution  of  x-ray  attenuation  coefficients 
from  transmitted  profiles  and  in  radiation  emis- 
sion tomography. 

4.    Computational  Procedure  and  Results 

As  presented,  the  problem  is  closer  to  that 
posed  in  emission  than  in  transmission  tomography. 
In  the  former  case  the  isotopic  distribution  is 
to  be  mapped  in  the  presence  of  attenuation.  In 
general  however  the  techniques  described  in  emis- 
sion tomography  either  ignore  the  effect  of  at- 
tenuation or  compensate  for  it  simply  by  comparing 
opposed  views.    In  general  no  serious  attempt  is 
made  to  generate  attenuation  maps  for  diagnostic 
use.    Clearly,  our  formulation  set  out  above  al- 
lows for  the  computation  of  both  back-scatter 
cross-sections  and  attenuation  coefficients 


simultaneously.  The  successful  implementation  of 
this  approach  is  described  later. 

A  variety  of  iterative  approaches  are  available 
and  are  under  investigation.    The  technique  which 
at  present  appears  most  promising  and  has  been  in- 
vestigated most  fully  generates  the  (n  +  l)th 
value  of  attenuation  coefficient  from  the  nth 


value  a"  by 


n+1 


W 


1  + 


Z(W2)  + 


V) 


(4) 


where  e  is  the  difference  between  the  data  value 
from  any  location  and  the  predicted  value  from  the 
current  image  arrays,  and  c  is  a  weighting  factor 
for  the  back-scatter  cross-section.    (WH  is  sum- 
med only  over  the  image  elements  in  the  beam  up  to 
the  position  related  to  the  data  value  used.  Simi- 
larly, the  (n  +  l)th  value  of  o.j  is  generated  from 
the  nth  value  using  eq.  (5) 


n+1 


(5) 


These  expressions  are  similar  to  those  used  by 
Gordon  et  al.  [7]  in  the  transmission  reconstruc- 
tion algorithms  ART. 

The  program  set  up  to  investigate  this  approach 
can  receive  either  simulated  or  real  data.  Simu- 
lation enables  up  to  10  circular  fields  of  speci- 
fied radius,  position  and  acoustic  properties  to 
be  used.    Figure  3  shows  a  reconstructed  image 
using  simulated  data  from  a  cylindrical  scattering 
region  containing  smaller  cylindrical  regions  of 
increased  attenuation.    The  larger  region  is  of 
12  mm  radius  and  has  an  attenuation  coefficient  of 


Fig.  3.    Reconstructed  attenuation  coefficient  map 
from  simulated  data  after  4  cycles.  The 
image  is  on  a  33  x  33  matrix  of  1  mm  square 
cells.    The  object  was  a  cylindrical  region 
12  mm  diameter,  attenuation  coefficient 
0.2  neper  cm"i  including  4  regions  4,  3, 
2,  1  mm  diameter,  attenuation  coefficient 
0.1  neper  cm"^.    Data  was  from  10  linear 
scans  with  pulse  spacing  0.5  mm.    The  as- 
sociated scatter  field  was  uniform. 


249 


0.2  neper  cm"i.    The  smaller  regions  have  radii 
4,  3,  2,  and  1  mm  and  attenuation  coefficients  of 
0.1  neper  cm"i.    The  image  shown  was  generated 
after  4  iterative  cycles,  i.e.  after  all  the  data 
had  been  used  4  times.    The  convergence  and  numeri- 
cal accuracy  of  the  computation  is  indicated  in 
figure  4.    The  attenuation  coefficient  values  in 
pixels  along  column  18  are  plotted  for  the  second, 
fourth  and  sixth  iterative  cycles.    There  is  little 
alteration  in  the  values  from  the  third  to  the 
fifth  cycle  and  the  solution  clearly  is  converging 
to  values  which  are  numerically  correct. 

Such  rapid  convergence  to  a  low  noise  image  is 
dependent  very  largely  upon  the  choice  of  iterative 
procedure  used.    In  particular  it  has  been  found 
that  the  most  rapid  convergence  has  been  from  a 
primary  scatter  map  generated  from  all  the  data 
either  averaged  or  peak  detected,  with  a  primary 
attenuation  map  set  to  zero.    Convergence  did  oc- 
cur from  a  primary  scatter  map  generated  from  a 
single  linear  scan,  but  was  very  slow  and  oscilla- 
tory in  pattern.    The  sequence  of  iterative  cor- 
rections has  also  been  observed  to  alter  the  noise 
characteristics  of  the  final  image.    Several  se- 
quence orders  have  been  investigated  including  se- 
quencing along  a  pulse  train,  in  a  forward  or  re- 


25 


30 


5        10)     15  20 
Image  cell 

Fig.  4.    The  convergence  of  attenuation  values  along 
column  18  of  figure  3  after  2,  4  and  6 
cycles  compared  with  the  simulated  object 
values.    The  starting  assumption  was  a  =  0 
everywhere.    Pulses  taken  in  sequence  and 
pulse  values  in  reverse  sequence. 


z!5 

Fig.  5.  Reconstructed  maps  of  backscatter  cross-section  (upper)  and  attenuation  coefficient  (lower) 
from  overlapping  regions.    The  objects  are  shown  on  the  left.    The  reconstructed  maps  with 
high  threshold  are  shown  in  the  center.    An  edge  artifact  appears  in  the  attenuation  map. 
Low  threshold  images  (right)  are  less  noisy  and  show  suppression  of  the  artifact. 

250 


verse  direction,  and  changing  order  in  which  pulses 
are  used.    An  additional  weighting  not  included  in 
eqs.  (4)  and  (5)  has  also  been  used  which  improved 
convergence,  to  compensate  for  the  fact  that  pixel 
values  close  to  the  transducer  are  modified  more 
often  than  those  at  a  distance  along  a  sampled 
pulse. 

Abrupt  changes  in  the  scatter  level  have  been 
found  to  result  in  edge  artifacts  occurring  in  the 
attenuating  field.    This  is  illustrated  in  figure 
5,  which  shows  reconstruction  of  a  3-component  sim- 
ulated field  with  overlapping  regions  of  altered  a 
and  a.    A  ring  artifact  in  the  attenuation  plot  is 
clearly  evident,  associated  with  the  perimeter  of 
the  small  scattering  region.    However,  when  a 
threshold  is  introduced  to  limit  the  conditions 
under  which  eqs.  (4)  and  (5)  operate  (as  in  fig.  5, 
right)  the  edge  artifact  is  suppressed.    It  is 
clear  that  in  tissue  it  is  unlikely  that  such  an 
abrupt  change  will  occur  but  such  a  threshold  also 
enables  a  control  to  be  kept  on  the  effect  of  high- 
ly directional  spectral  reflections. 

5.  Discussion 

A  convergent  iterative  computational  procedure 
has  been  developed  to  solve  the  set  of  equations 
describing  quantitatively  the  back-scatter  of 
sound  from  a  distributed  scattering  and  attenuat- 
ing medium.    It  has  been  shown  that,  under  simula- 
tion conditions,  accurate  reconstructed  maps  of 
both  back-scatter  and  attenuation  distributions 
can  be  obtained.    Some  problems  posed  by  the  ap- 
plication of  the  techniques  to  real  data  have  al- 
ready been  investigated.    Since  an  iterative  pro- 
cedure is  being  used,  it  is  possible  to  include 
controls  which  limit  the  range  of  application  with- 
in the  iterative  process  to  include  only  those 
which  are  deemed  to  force  the  process  to  converge 
effectively.    It  is  clear  however  that  the  success 
of  the  application  in  tissue  will  depend  upon  the 
extent  of  validity  of  some  of  the  assumptions  used 
in  the  model,  and  this  is  yet  to  be  evaluated. 
Anisotropy  in  o  or  a,  position  errors  caused  by 
refractions  and  the  effects  of  a  finite  beam  width 
and  diffraction  errors  must  all  place  some  ulti- 
mate limitation  on  the  method.    Within  these  limi- 
tations however  the  techniques  offer  the  potential 
of  a  quantitative  imaging  of  back-scatter  cross- 
section  and  attenuation  coefficients  from  pulse- 
echo  data. 


6.  References 

[1]    McDicken,  W.  N.,  Evans,  D.  H.,  and  Robertson, 
D.  A.  R. ,  Automatic  sensitivity  control  in 
diagnostic  ultrasonics.  Ultrasonics  12,  173- 
176  (1974). 

[2]    DeClercq,  A.  and  Maginness,  M.  G.  Adaptive 
Gain  Control  for  Dynamic  Ultrasound  Imaging, 
in  IEEE  Ultrasonics  Symposium  Proceedings 
pp.  59-63  (1975). 

[3]    Greenleaf,  J.  F.,  Johnson,  S.  A.,  Lee,  S.  L., 
Herman,  G.  T.,  and  Wood,  E.  H.,  Algebraic 
Reconstruction  of  Spatial  Distributions  of 
Acoustic  Absorption  within  Tissue  from  Their 
Two-Dimensional  Acoustic  Projections,  in 
Acoustical  Holography,  P.  S.  Green    ed.,  vol. 
5. ,  p.  591  (Plenum  Press,  New  York,  1974). 

[4]    Greenleaf,  J.  F.  and  Johnson,  S.  A., 

Algebraic  Reconstruction  of  Spatial  Distri- 
butions of  Refractive  Index  and  Attenuation 
in  Tissues  from  Time-of-Fl ight  and  Profiles, 
in  Ultrasonic  Tissue  Characterisation,  M. 
Linzer  ed..  National  Bureau  of  Standards 
Spec.  Publ.  453,  p.  109  (U.S.  Government 
Printing  Office,  Washington,  D.C.,  1976). 

[5]    Radon,  J.,  Uber  die  Bestimmung  von  Funktionen 
durch  ihre  Integralwerte  langs  gewisser 
Mannigfaltigkeiten.    (On  the  determination 
of  functions  from  their  integrals  along  cer- 
tain manifolds.)    Berichte  Sachsische  Aka- 
demie  der  Wissenschaften  (Leipzig),  Mathe- 
matische-Physische  Klasse  69.  262-277  (1917). 

[6]    Gordon,  R.  and  Herman,  G.  T. ,  Three-dimension- 
al reconstruction  from  projections:    a  review 
of  algorithms.  International  Review  of  Cytolo- 
gy. 38  ,  111-151  (1974).   

[7]    Gordon,  R. ,  A  tutorial  on  ART,  IEEE  Trans. 
Nuc.  Sci. .  NS-21,  78-93  (1974). 


251 


CHAPTER  9 

SIGNAL  PROCESSING  AND  PATTERN  RECOGNITION 


253 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


A  COMPREHENSIVE  ULTRASONIC  TISSUE  ANALYSIS  SYSTEM 

M.  Linzer,  S.  I.  Parks,  S.  J.  Norton, 
F.  P.  Higgins',  D.  R.  Dietz^  and  R.  W.  Shideler 

National  Measurement  Laboratory 
National  Bureau  of  Standards 
Washington,  D.C.  20234 

and 

T.  H.  Shawker  and  J.  L.  Doppman 

Clinical  Center 
National  Institutes  of  Health 
Bethesda,  Maryland  20014 

A  progress  report  on  the  development  of  a  comprehensive  system  for  ultrasonic 
tissue  characterization  is  presented.    Major  elements  of  the  program  include  com- 
puterized tomography  studies,  particularly  for  breast  cancer  detection;  opto-acoustic 
visualization  of  ultrasonic  fields,  for  testing  of  new  imaging  schemes,  studies  of 
propagation  through  inhomogeneous  media,  in  vitro  measurements,  and  transducer  cali- 
bration; electronic  focusing,  especially  annular  array  imaging;  sensitivity  enhance- 
ment, using  digital  signal  averaging  and  pulse  compression  techniques;  computer  and 
chirp  waveform  techniques  for  compensation  of  frequency-dependent  attenuation;  the 
SonoChromascope,  a  digital  device  for  real-time  acquisition,  processing,  and  display 
of  B-scan  images;  and  computer-based  image  processing. 

Key  words:    Annular  array;  breast  cancer;  chirp  signals;  imaging;  opto-acoustic; 

pulse  compression;  sensitivity;  signal  averaging;  signal  processing; 
tissue  characterization;  tomography;  transducers;  ultrasonics. 


I.  INTRODUCTION 

This  paper  reports  on  the  progress  of  a  Com- 
prehensive Ultrasonic  Tissue  Analysis  System 
(CUTAS)  which  is  being  developed  by  the  Signal 
Processing  and  Imaging  Group,  Center  for  Materials 
Science,  NBS,  and  clinically-evaluated  by  the 
Department  of  Diagnostic  Radiology,  Clinical 
Center,  NIH.    Three  principal  areas  are  being 
emphasized  in  this  work: 

1.  Computerized  tomography  studies,  particularly 
for  breast  imaging, 

2.  Optical  visualization  of  ultrasonic  fields. 

3.  General  purpose  clinical  studies. 

II.    COMPUTERIZED  TOMOGRAPHY  STUDIES 

Major  emphasis  to  date  has  been  the  develop- 
ment of  backprojection  algorithms  for  reconstruc- 
tion of  cross-sectional  images  of  reflectivity 
using  a  circular  array  of  transducer  elements 
enclosing  the  object.    Three  basic  modes  of  data 
acquisition  and  image  reconstruction  were 
analyzed  (fig.  1):    (1)  the  same  element  serves 
as  transmitter  and  receiver  and  data  is  backpro- 
jected  along  circular  paths  centered  at  the 


'Present  address:    Western  Electric  Company, 
Princeton,  New  Jersey 

^Present  address:  McDonnell  Douglas  Corporation, 
St.  Louis,  Missouri 


0  2p/c  0  (P,+p,l/c 


Fig.  1.    (a)  Operation  of  circular  array  with 
same  element  serving  as  transmitter 
and  receiver.    Each  point  in  the 
A-scan  is  the  sume  of  echoes  arising 
from  scatterers  along  a  circular  arc 
centered  at  the  active  element,  (b) 
Operation  of  circular  array  with  sepa- 
rate transmitter  and  receiver  elements. 
Each  point  in  the  A-scan  is  the  sum  of 
echoes  arising  from  points  lying  along 
an  elliptical  arc  whose  foci  are  the 
transmitter  and  receiver. 

element;  (2)  distinct  transmitter  and  receiver 
with  fixed  separation  and  backprojection  along 
elliptical  paths  with  the  elements  at  the  foci; 
and  (3)  distinct  transmitter  and  receiver  with 
varying  separations  and  backprojection  along 
corresponding  elliptical  paths.    Point  spread 
functions  (PSF's)  were  evaluated  for  narrowband, 
wideband,  and  an  analytically-derived  optimum 
pulse  which  yields  the  best  sidelobe  response 
and  a  mainlobe  width  equal  to  one-third  of  the 


255 


1.00 


0.75 


0.50 


1  0.25 


-0.25 


— - — si'Nyx/y  ,  y\/\,^Nv  ^ 


-4 


-2       0  2 
r  Imm) 


Fig.  2.    Point  spread  function  obtained  by  back- 
projecting  an  analytically-derived 
optimum  pulse  along  circular  paths  in 
image  space. 


including  that  of  the  transducer  face.    The  ultra- 
sonic wavefronts  at  other  planes  were  also 
measured  and  compared  to  the  calculated  values. 
Both  narrowband  (gated-cw)  and  wideband  (pulsed) 
waveforms  were  used  to  excite  the  transducer. 
Computerized  holographic  reconstructions  of 
several  model  targets  were  also  made.    An  example 
of  a  reconstructed  plane  in  the  field  of  a  pulsed 
transducer  is  shown  in  figure  3.    Other  planned 


NORMALIZED 
DISPLACEMEfjT 
AMPLITUtSE 


96  mm  X  6  4  ps 


96  mm  X  6,4  ps 


wavelength  corresponding  to  the  cut-off  frequency 
of  the  pulse  (fig.  2).    When  backprojection  is 
performed  along  elliptical  paths,  corresponding 
to  two  separate  elements,  the  mainlobe  of  the  PSF 
was  shown  to  be  broadened  by  a  factor  proportional 
to  the  cosine  of  half  the  angle  subtending  the 
two  elements  at  the  center  of  the  array.  This 
behavior  places  a  practical  restriction  of  about 
45°  on  the  angular  separation  of  the  elements. 
Computer  simulations  confirmed  the  salient 
properties  predicted  by  the  analytically-derived 
PSF's.    The  characteristics  of  the  PSF's  were 
calculated  as  a  function  of  the  number  of  array 
elements,  the  position  of  the  reflecting  point 
within  the  object,  and  the  shape  of  the  pulse. 

The  backprojection  analysis  has  been  extended 
to  three-dimensional  reconstructions  (using 
spherical  transducer  arrays)  and  to  incorporate 
corrections  for  velocity  inhomogeneities  in  the 
medium.    Work  in  progress  includes  the  develop- 
ment of  a  perturbation  approach  for  correcting 
velocity  images  and  algorithms  for  frequency- 
dependent  time-gain  compensation.    A  prototype 
breast  scanner  is  now  under  construction  and  will 
be  used  to  evaluate  these  various  developments 
in  ultrasound  tomography. 

III.  OPTICAL  VISUALIZATION  OF  ULTRASONIC  FIELDS 

A  major  laboratory  facility  for  reconstruction 
and  optical  visualization  of  ultrasonic  fields 
has  been  developed.    The  system  is  based  on  the 
use  of  a  Michel  son  interferometric  system  which 
is  capable  of  measuring  the  displacement  of  a 
thin  metallized  150  mm  diameter  pellicle  immersed 
in  water.    Major  advances  of  the  NBS  development 
over  previous  designs  include  measurement  of  both 
the  amplitude  and  phase  of  the  displacement,  and 
absolute  calibration  of  the  displacement  to 
approximately  one  percent  accuracy. 

This  capability  of  measuring  the  complex  ultra- 
sound field  over  a  large  aperture,  coupled  with 
computer  acquisition  and  analysis  of  the  experi- 
mental data,  has  provided  us  with  a  powerful 
research  tool  for  visualization  and  reconstruction 
of  ultrasonic  wavefronts.    In  our  initial  work, 
the  ultrasonic  field  of  a  transducer  was  measured 
at  the  plane  of  the  pellicle  placed  near  the 
transducer  face.    The  exact  (plane  wave)  solution 
was  used  to  reconstruct  the  field  at  other  planes. 


Fig.  3.    (a)  Computer  reconstruction  of  a  propagat 
ing  wideband  pulse  at  100  mm  (far-field) 
from  transducer  face.    Reconstruction  was 
based  on  measurement  of  the  time-depend- 
ence of  the  amplitude  and  phase  of  the 
pulse  in  a  near-field  plane,    (b)  Actual 
measurement  of  pulse  at  100  mm  plane. 

uses  of  the  system  include  studies  of  new  ultra- 
sonic imaging  schemes,  measurements  of  ultrasonic 
parameters  of  tissue  in  vitro,  and  examination  of 
ultrasonic  propagation  through  inhomogeneous 
media. 

IV.  GENERAL-PURPOSE  CLINICAL  STUDIES 

A  general -purpose  clinical  scanner  is  now  under 
active  development  (fig.  4).    Major  components 
of  the  system  include  annular  array  focusing  for 
improving  resolution;  sensitivity  enhancement 
techniques  using  A-  and  B-scan  averaging,  chirp- 
radar  approaches  and  focusing;  first-order 
corrections  for  frequency-dependent  attenuation 
in  the  medium;  and  improved  signal  processing 
approaches  combining  the  real-time  acquisition, 
processing,  and  display  capability  of  the  Sono- 
Chromascope  with  computer  post-processing.  To 
date,  our  approach  has  been  to  evaluate  each 
component  separately  as  it  is  developed.  Even- 
tually, we  expect  to  integrate  these  various 
elements  into  a  unified  facility  for  clinical 
ultrasound  examinations. 

A.    Annular  Array 

1.    Expanding-Aperture  Annular  Array 

A  dynamically-focused  annular  array  system  for 
contact  B-scanning  has  been  developed  (fig.  5). 
The  design  is  based  on  a  constant  F-number 
approach,  whereby,  at  short  focal  lengths,  the 
aperture  is  expanded  in  proportion  to  the  focal 
length.    This  approach  allows  the  use  of  wider 
array  elements,  thus  increasing  the  sensitivity 
of  the  system  and  reducing  radial -mode  coupling. 
Other  major  advantages  include  a  substantial 
reduction  in  the  delays  and  refocusing  rates 
required  for  the  lens  synthesis  with  a  correspon- 
ding reduction  in  the  electronic  complexity  of 
the  system.    The  initial  design  employs  an  array. 


256 


ULTBAFAST 

SIGNAL 
ENHANCER 


A  SCAN  DIGITIZATION 
AND  AVERAGING 


ANNULAR 

ARRAY 
SCANNER 


RFSUBSYSTEM 


PREPROCESSING 


SONOCHROMASCOPE 


MULTIPARAMETER 
ACQUISITION  PROCESSING 
AND  DISPLAY 


NONLINEAR 
STUDIES 


VIDEOTAPE 

A  SCAN 
RECORDER 


Fig.  4.      Block  diagram  of  general  purpose  clinical  ultrasound  system. 


FREQUENCY  25MHi 
DIAMETER  3  9  CM 

MUMBER  OF  RINGS  12 


Fig.  5.      Operation  of  the  expanding-aperture 

array  with  dynamic  focusing  on  receive. 

operating  at  2.25  MHz,  with  four  annul i  active 
at  the  near  focal  length  of  1.5  cm.    As  the  focal 
length  increases,  the  array  expands  to  a  maximum 
of  twelve  rings,  with  4.0  cm  outer  diameter,  for 
focal  lengths  greater  than  12  cm.    A  single, 
tapped  delay  line  with  1  \is  total  duration  pro- 
vides the  time  delays  for  focusing  on  receive. 
A  continuously-variable  point  or  line  focus  is 
provided  on  transmit.    Experimental  measurements 
vides  the  time  delays  for  focusing  on  receive. 
A  continuously-variable  point  or  line  focus  is 
provided  on  transmit.    Experimental  measurements 
of  the  focusing  properties  of  the  system  have  been 
made.    Some  of  these  results  are  shown  in  figures 
6  and  7.    Clinical  evaluation  of  the  annular  array 
scanner  is  now  underway. 


Fig.  6.    B-scans  of  AIUM  test  object  with  a  single 
fixed-focus  transducer  (a)  and  with  the 
dynamically-focused  annular  array  (b). 


> 


 1  1  1 — 

'  A ' 

-r        I         1  ' 

Uni 

:rary 

(Arbil 

/  ^■'^  \ 

j   mm  \ 

IGNAL 

.1       1       1       1       1       1  1 

-8    -6  -4 

-2     0  2 

4     6  8 

OFF-AXIS  DISTANCE  IN  FOCAL  PLANE  (mm) 

Fig.  7.      Measured  response  of  the  annular  array 

focused  at  10  cm  on  transmit  and  receive. 

2.    Wideband  Annular  Array  Response 

Theoretical  studies  of  the  effect  of  bandwidth 
on  the  focal  plane  response  of  a  circular  lens 
and  annular  array  were  carried  out.  Particular 
emphasis  was  placed  on  lens  systems  operating  at 
approximately  50  percent  bandwidth,  typical  of 
those  used  in  ultrasound  imaging.    An  analytical 
model  of  the  focal  plane  response  of  both  the 
circular  lens  and  annulus,  driven  by  an  impulse, 
was  developed.    The  wideband  response  was  then 
calculated  by  convolving  the  impulse  response 
with  the  driving  function.    For  a  circular  lens, 
the  beam  width  in  the  focal  plane,  as  well  as  the 
position  and  height  of  the  sidelobes,  was  analyzed 
as  a  function  of  bandwidth  and  aperture  weighting. 
The  wide  bandwidth  model  of  an  annulus  was  used  to 
calculate  the  response  of  an  annular  array.  A 
detailed  comparison  was  made  of  this  model  with 
the  experimentally-measured  response  of  an  array, 
operating  at  2.25  MHz  with  40  percent  bandwidth 
(fig.  8). 

B.  Ultrafast  Signal  Averaging  and  Pulse 
Compression  Techniques 

A  signal  averager  and  pulse  compression  system 
has  been  developed  for  sensitivity  enhancement 
in  ultrasonic  diagnosis.    Potential  applications 
include  the  use  of  transducers  which  are  ineffi- 


257 


-8      -6      -4      -2        0        2        4        6  8 


OFF-AXIS  DISTANCE  p  (mm) 


Fig.  9.      Ultrasonic  reflection  from  posterior 
cortex  of  tibia;    (a)  no  averaging; 
(b)  after  averaging  for  scans. 


Fig.  8.    Field  pattern  at  10  cm  generated  by  a 
0.4  diameter  piston  transmitter  and  a 
thin,  1.4  cm  diameter,  annulus  receiver. 
The  solid  line  is  the  calculated 
response  and  the  dashed  line  is  the 
measured  response. 

cient  but  otherwise  have  very  desirable  features 
(e.g. ,  point  and  line  sources,  polymer  trans- 
ducers, CdS  phase-insensitive  transducers); 
higher  frequency  operation  for  improved  pattern 
recognition  and  increased  resolution;  detection 
of  small  reflections;  examination  of  obese 
patients;  and  penetration  through  skull  and  bone. 
These  systems  also  make  it  possible  to  reduce 
peak  power  while  keeping  average  power,  and  hence 
sensitivity,  constant. 

1 .  Signal  Averager 

The  signal  averager  is  capable  of  real-time 
(unbuffered)  averaging  at  50  MHz  rates.    To  our 
knowledge,  this  device  is  the  fastest  digital 
averaging  device  in  existence.    Major  features 
include  4K  24-bit  words,  12.5  kHz  maximum  repeti- 
tion rate,  computer  interface,  6-digit  cursor 
readout  of  signal  amplitude,  region  of  interest 
expansion  (up  to  a  factor  of  16),  3-digit  setta- 
bility  of  sample  rate,  internal/external  trigger, 
internal  delay,  segmented  memory  capability  (full, 
halves,  quadrants,  octants),  plug-in  ADC's  (4  bit, 
50  MHz;  8  bit,  20  MHz),  ADC  resolution  enhancement 
(via_  ordered  dither),  display  normalization  and 
semi-real  time  display  at  high  frequencies.  The 
sensitivity-enhancement  capability  of  the  averager 
was  demonstrated  on  a  PZT  line  source  and  on  a 
number  of  highly-attenuating  biomaterial s,  includ- 
ing human  tibia  (fig.  9),  and  a  fiber  composite 
(fig.  10)  used  in  synthetic  implants. 

2.  Pulse  Compression 

The  pulse  compression  circuit  incorporates  a 
surface  acoustic  wave  (SAW)  "chirp"  filter.  Pulse 
compression  ratios  of  30:1  and  8:1  have  been 
obtained  in  the  case  of  8  MHz  and  3  MHz  filter 
bandwidths,  respectively.    An  example  of  both  an 
expanded  and  compressed  (8:1  compression  ratio) 
echo  from  a  human  heart  in  vivo  is  shown  in 
figure  11.    The  chirp  system  was  also  used  to 
compensate  for  frequency-dependent  attenuation  in 
the  medium  by  modulating  the  amplitude  of  the 


(a) 


(b) 

Fig.  10.  Backwall  reflection  from  5  cm  thick 
fiber  composite:  (a)  no  averaging; 
(b)  after  averaging  for  2^^  scans. 


Fig.  n.    Preamplified  echo  from  a  human  heart  in 
vivo.    The  initial  transient  is  the 
clipped  transmission  pulse;  the  remaining 
echoes  are  from  tissue:    (a)  expanded 
pulse;  (b)  pulse  after  8:1  compression. 

expanded  stimulus  pulse  as  a  function  of  time. 
Since  the  frequency  of  this  pulse  was  also  time- 
dependent,  the  strongly-attenuated  frequencies 
were  enhanced.    This  approach  has  important  impli- 
cations for  improving  range  resolution  and  for 
tissue  characterization  and  is  now  under  detailed 
investigation. 


258 


C.  SonoChromascope 

The  SonoChromascope  (fig.  12)  is  a  state-of- 
the-art  device  for  the  digital  acquisition, 
processing,  recording,  and  display  of  ultrasonic 
B-scaii  images. 


Fig.  12.    Clinical  ultrasound  examination  using 
the  SonoChromascope. 


imaging,  including  electronic  focusing  and  com- 
puterized tomography;  sensitivity  enhancement; 
measurement  of  tissue  parameters  in  vitro; 
studies  of  ultrasound  propagation  through  inhomog- 
eneous  media;  computer  and  chirp  techniques  for 
compensation  of  frequency-dependent  attenuation; 
real-time  digital  processing  and  display  tech- 
niques; and  computer-based  image  processing. 
Several  of  these  developments  have  already  been 
completed,  and,  v/here  appropriate,  are  undergoing 
clinical  testing.    Complete  descriptions  of  this 
work  will  be  published  shortly. 


Acquisition  algorithms  include  a  choice  of 
log  or  linear  detection;  recording  the  minimum 
and/or  maximum  echo  for  each  pass  through  a  pixel 
(thus,  for  example,  providing  a  measure  of 
scattering  anisotropy);  unconditionally  writing 
the  last  value;  and  summing  with  normalization 
(thereby  improving  sensitivity).    A  combination 
of  up  to  four  different  algorithms  may  be  applied 
simultaneously  to  produce  different  spatially- 
congruent  images  during  the  same  B-scan. 

After  acquisition,  the  image  or  images  are 
displayed  on  two  color-TV  monitors.    Images  in 
complementary  colors  may  be  overlaid  to  permit 
visual  discrimination  of  subtle  differences.  A 
variety  of  thresholding  and  display  modes,  aug- 
mented by  a  lightpen,  permit  semiquantitative 
measurements  of  ultrasound  echo  intensity.  To 
use  the  lightpen,  the  operator  "paints  in"  the 
area  of  interest.    As  he  does  so,  the  average 
image  intensity  (or  intensities)  and  the  area 
painted  in  are  displayed  on  digital  readouts. 

If  more  sophisticated  processing  or  storage 
is  required,  the  data  may  be  transferred  to  a 
minicomputer.    Processed  or  stored  images  may  be 
returned  to  the  SonoChromascope  for  display. 

Spatial  and  amplitude  resolution  depend  upon 
the  acquisition  mode.    For  example,  in  the  "un- 
conditional write"  mode,  the  word  size  is  8  bits 
and  the  picture  contains  480  x  480  pixels;  in 
the  sum  mode,  the  word  size  is  14  bits  and  there 
are  480  x  240  pixels. 

The  SonoChromascope  is  presently  interfaced 
to  a  commercial  B-scan  system,  from  which  it 
receives  the  rf^or-  log -detected  A-scan  signal 
and  appropriate  information  about  transducer 
position.    It  is  now  undergoing  clinical  evalua- 
tion. 

V.  SUMMA.RY 

A  major  effort  is  now  underway  in  developing 
a  comprehensive  system  for  ultrasonic  tissue 
characterization.    The  program  encompasses 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


THEORETICAL  ANALYSIS  OF  INSTANTANEOUS  POWER  SPECTRA 
AS  APPLIED  TO  SPECTRA-COLOR  ULTRASONOGRAPHY 


W.  D.  Jennings,  E.  Holasek,  and  E.  W.  Purnell 

Department  of  Surgery,  Division  of  Ophthalmology 
School  of  Medicine 
Case  Western  Reserve  University 
Cleveland,  Ohio    44106,  U.S.A. 


Spectra-color  ultrasonography  (SCU),  a  technique  for  two-dimensional  (B-mode)  dis- 
play of  ultrasonic  spectral  data,  has  been  analyzed  theoretically.    The  analysis  in- 
cludes the  use  of  an  instantaneous  power  spectrum  calculated  from  broadband  gated  echo 
spectra  produced  by  an  analog  spectrum  analyzer.    The  results  of  the  analysis  indicate 
that  additional  signal  processing  factors  must  be  added  to  the  SCU  system  as  it  was 
originally  designed.    With  the  inclusion  of  these  modifications,  the  SCU  scan  rep- 
resents a  true  low  resolution  instantaneous  spectral  analysis  of  ultrasonic  echo  wave- 
forms.   An  experiment  was  performed  to  test  the  theoretical  equations  we  have  develop- 
ed relating  SCU  to  an  instantaneous  spectral  analysis.    The  comparison  of  the  SCU 
signals  and  the  computed  SCU  equivalent  based  on  an  instantaneous  power  spectrum  is 
presented. 


Key  words:    Instantaneous  power  spectra:    Color-coded  B-scan;  spectra-color  ultra- 
sonography (SCU);  spectrum  analysis;  ultrasonic  spectroscopy. 


1.  Introduction 

Several  years  ago  we  developed  spectra-color 
ultrasonography  (SCU)  which  was  originally  de- 
signed as  an  approximate  spectroscopic  technique 
to  display  two-dimensional  (B-mode)  spectroscopic 
information  from  a  sector  scanned  tissue  segment 
[l,2]i.    Originally,  the  SCU  image  was  formed  by 
first  passing  a  broadband  B-scan  ultrasonic  echo 
train  through  a  voltage  tuned  filter  set  at  three 
different  center  frequencies  (bandpass  positions). 
Each  of  the  three  resultant  filtered  B-scans  was 
then  assigned  a  color  (red,  green  or  blue).  The 
three  colored  B-scans  were  then  superimposed  to 
form  a  single  frequency-dependent  color  coded 
B-scan  display.    The  final  colored  B-mode  image 
displayed  frequency-dependent  properties  of  the 
tissue  being  scanned,  though  the  precise  mean- 
ing of  the  display  was  not  apparent. 

The  subject  of  this  presentation  is  an  analysis 
of  the  SCU  process  in  terms  of  an  instantaneous 
power  spectrum.    An  instantaneous  power  spectrum 
is  a  time  dependent  spectral  decomposition  of  a 
signal.    The  conventional  Fourier  Power  Spectrum 
of  a  signal  is  defined  by  an  integral  of  the 
signal  over  all  time,  and  thus,  is  independent  of 
time.    An  instantaneous  power  spectrum,  however, 
is  a  function  of  both  time  and  frequency.    It  is 
the  dual  functionality  of  instantaneous  power 
spectra  that  makes  them  applicable  to  signals  that 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


originate  from  spatially  distributed  sources  with 
varying  spectral  content,  such  as  ultrasonic  echo 
waveforms. 

Several  definitions  of  an  instantaneous  power 
spectrum  have  been  devised.    We  began  our  work  on 
instantaneous  power  spectra  by  attempting  to  adapt 
a  particular  instantaneous  power  spectrum,  the  one 
defined  by  Chester  Page  [3],  to  spectra-color 
ultrasonography.    During  our  investigations,  ex- 
perimental evidence  led  us  to  the  conclusion  that 
for  spectra-color  ultrasonography  to  represent  a 
true  instantaneous  power  spectrum,  it  is  appro- 
priate to  use  Morris  Levin's  modification  of 
Page's  definition  of  an  instantaneous  power  spec- 
trum, rather  than  Page's  original  definition. 
This  paper  is  an  account  of  those  investigations. 

2.  Theory 

Essential  to  the  analysis  of  spectra-color 
ultrasonography,  is  the  concept  of  an  instantaneous 
power  spectrum.    We  began  with  the  approach  of 
Chester  Page  [3]  which  we  will  briefly  review  here. 
For  any  time  domain  signal  g(t)2,  the  running 
Fourier  transform  of  g(t)  is  defined  as 

t 

G^(co)  =   f  g(x)e-J'^^dx    .  (1) 


2ln  our  case,  g(t)  is  the  receiving  transducer 
voltage  corresponding  to  the  instantaneous 
acoustic  pressure  on  the  transducer  face. 


261 


That  is,  G^((;))  is  the  Fourier  transform  of  the 
semi-infimte  signal  g(x)  taken  from  x  =  -«>  to 
X  =  t.    Note  here  that  x  is  a  dummy  variable,  al- 
lowing us  to  use  the  time  variable  t  as  the  trail- 
ing edge  of  the  gate  on  g(t).    The  instantaneous 
power  spectrum  of  g(t)  can  then  be  defined  as 


p(t,co)  = 


3t 


|G^(a)) 


Page  also  shows  that    p(t,to)  can  be  represented  by 


p(t,co)  =  2g(t)Re 


Jwt 


xle-J-^^dx 


(3) 


By  combining  the  exponential  terms,  and  interchang- 
ing the  order  of  integration  we  get 


J.(t)  =  2g(t)Re 


(2)  or 


Ji(t)  =  2g(t)Re 


J  j  C.(a))g(x)eJ'^'^"'')da)dx 


/  g(x)/  C.(a,)ej'^(^-^)da)dx 


(9) 


(10) 


By  use  of  eq.  (4),  eq.  (10)  becomes 


Using  eq.  (3)  as  a  starting  point,  we  have  ana- 
lyzed SCU  as  follows:  In  the  original  SCU  implemen- 
tation, the  signal  g(t)  was  passed  through  a  voltage 
tuned  filter  set  at  three  different  bandpass  posi- 
tions.   In  the  frequency  domain,  the  i**^  filter  can 
be  described  by  some  function  C-j(u).    In  general, 
Ci(u))  is  a  complex  function,  including  phase  delays 
introduced  by  the  filter  as  a  function  of  frequency. 
The  time  domain  representation  of  the  filter  will  be 
designated  by  c-j(t),  which  is  also  complex  in  general 
The  function  c-j(t),  is  the  inverse  Fourier  transform 
of  C-j(u),  and  is  the  impulse  response  of  the  filter 


J.(t)  =  4TTg(t)Re 


J  g(x)c.(t- 


x)dx 


(11) 


For  a  filter  that 
(ci(t)  =  0  for  t 


is  causal  in  the  time  domain 
:  0) ,  eq.  (11)  becomes 


Ji(t) 


4TTg(t)Re 


j  g(x)c.(t-x)dx 


(12) 


:.(t)  =  J-/c.(.)eJ 


jwt 


dto 


(4) 


Thus,  the  filtered  signal  can  be  described  in  the 
time  domain  by 


S.(t)  =  Re[c.(t)*g(t)j 


(5) 


What  we  would  like  to  display  is  the  band-limited 
instantaneous  power  spectrum  of  the  signal.  That 
is,  the  desired  display  is 


Ji(t)  =   j  p(t,w)C.(aj)dco 


(6) 


Ji(t)  is  the  instantaneous  power  (as  defined  by 
Page)  of  the  signal  g(t)  in  the  frequency  band  de- 
scribed by  C-i(a)).    Substituting  eq.  (3)  into  eq. 
(6)  and  rearranging  we  get 


J.(t)  =  2g(t)  j  C.(w)Re  e^^^^y  g(x)e"J^^dx 


dw  .  (7) 


If  we  assume  at  this  point  that  the  filter  function 
Ci(u))  is  a  real  valued  function,  rather  than  com- 
plex, we  can  write 


J.(t)  =  2g(t)Re 


or 


J.(t)  =  2g(t)Re 


/ 


C.(a3)e 


jut 


/g(^ 


x)e  '^'^^dxdw 


(8) 


j  j  C.(a))eJ'^S(x)e"^'''dxdw 


The  integral  in  eq.  (12)  is  the  convolution  in- 
tegral for  g(t)  and  c-j(t).    Using  the  notation  of 
eq.  (5),  eq.  (12)  becomes 


J^(t)  =  4^g(t)S.(t) 


(13) 


Thus,  we  come  to  the  conclusion  that  for  SCU  to 
represent  a  true  color  coded  low  resolution  in- 
stantaneous spectral  analysis  we  must  modify  the 
original  system  in  two  ways: 

(1)  The  filters  used  must  have  filter  functions 
that  are  real  in  the  frequency  domain,  or  be 
modified  so  as  to  be  effectively  real. 

(2)  The  filtered  waveforms  must  be  multiplied 
by  the  broadband  signal  before  video  processing 
and  display. 

3.    Experimental  Approach 

We  have  tested  the  theory  by  comparing  J-j(t) 
calculated  by  eq.  (6)  with  J-j(t)  defined  by  eq. 
(13),  which  mathematically  represents  the  modified 
SCU  system. 

Calculation  of  J-j(t)  by  eq.  (6)  requires  mea- 
surement of  the  filter  functions  C-j(a)).    We  used 
three  fixed  filters  with  1.6  MHz  bandwidths  center- 
ed at  7,  9,  and  11  MHz.    The  fixed  filters  re- 
placed the  voltage  tuned  filter  used  in  earlier 
implementations  of  spectra-color  ultrasonography. 
The  spectrum  of  the  ultrasonic  echo  from  a  glass 
block  was  used  as  a  reference  spectrum,  and  the 
spectrum  of  the  signal  passed  through  each  filter 
was  recorded.    The  filtered  spectra  and  reference 
are  shown  in  figure  1.    The  spectra  are  plotted 
from  5  to  15  MHz.    A  5  MHz  crystal  controlled 
oscillator  was  used  to  provide  a  frequency  marker 
for  accurate  spectral  measurements.    All  spectra 
were  recorded  by  our  automated  data  acquisition 


262 


wideband 


9  10  11  12  13 
Frequency  (MHz) 


14 


Fig. 


1.  Wideband  reference  spectrum  and  spectra  of 
the  low,  middle  and  hiqh  bandpass  filtered 
reference  signals. 

system  [4]  controlled  by  a  programmable  calculator. 

A  natural  sponge  was  used  as  the  target  mate- 
rial because  of  its  stability  and  similarity  to 
human  tissue  in  acoustical  properties.    The  in- 
stantaneous power  spectrum,  p(t,a))  required  by 
eq.  (6),  was  calculated  by  eq.  (2).    To  produce 
the  semi-infinite  signal  gt(x)  we  constructed  a 
programmable  gating  system.    Using  the  crystal 
controlled  5  MHz  source  as  a  clock,  the  gate  could 
be  automatically  varied  in  length  in  200  nano- 
second increments.    Power  spectra  |G^((o)|2  of  the 
signal  gated  at  increasing  lengths  were  recorded. 
The  value  of  (3/3t) |Gt(uj)|2  was  calculated  by  sub- 
tracting spectra  of  gated  signals  of  increasing 
length  to  approximate  the  differential.  Ji(t) 
was  then  calculated  as  the  summation  of 
Ci(io)  •  (8/9^.)  |Gt(a))  |2  over  all  frequencies  to  ap- 
proximate the  integral  in  eq.  (6). 

To  determine  Ji(t)  by  eq.  (13),  the  filtered 
signal,  Si(t),  was  multiplied  by  the  broadband 
signal,  g(t),  using  a  double  balanced  modulator. 
A  broadband  pulse  amplifier  was  used  to  bring  the 
level  of  g(t)  into  the  operating  range  (approxi- 
mately one  volt)  of  the  modulator.    The  modulator 
provided  bipolar  multiplication  of  the  two  signals, 
as  shown  in  figure  2,    The  multiplied  signal  for 
each  of  the  three  bandpass  filters  is  shown  in 
figure  3. 

For  each  filter,  an  appropriate  precision  de- 
lay line  was  added  to  the  broadband  signal  before 
multiplication  in  the  modulator.    The  delay  lines 


Fig.  2. 


2c- 


Analog  signal  multiplication  for  the 
middle  frequency  band,    a)  multiplied 
signal,  J(t);  b)  wideband  signal,  g(t): 
c)  filtered  waveform,  Si(t). 


3a 


3b 


u 

i 

3c 


Fig.  3.    Analog  multiplied  signals,  J^(t)  fot  the 
three  frequency  bands,    a)  high  frequency 
band;  b)  middle  frequency  band;  c)  low 
frequency  band. 

were  necessary  to  compensate  for  the  group  delay 
through  the  filters.    Group  delays  were  approxi- 
mately 1  microsecond,  but  were  different  for  each 
filter.    Thus,  the  requirement  that  the  filter 
functions  be  real  was  approximately  satisfied. 

4.    Results  and  Discussion 

Figures  4a,  4b,  and  4c,  show  J-i(t)  calculated  by 
eq.  (6),  and  J-j(t)  determined  by  the  signal  multi- 
plication described  in  eq.  (13)  for  the  three 
filters.    The  time  resolution  of  the  calculated 
signals  is  one  point  every  200  nanoseconds,  while 
the  analog  multiplied  J^{t)  is  a  continuous  func- 
tion of  time.    The  calculated  Ji(t)  represents  the 
average  of  J-j(t)  over  the  200  nanosecond  interval. 

Noise  in  the  system  was  reduced  by  averaging  in 
the  case  of  the  calculated  Ji(t).    The  plots  shown 
in  figure  4,  represent  an  average  of  40  scans 
through  the  5  microsecond  section  of  sponge.  Fig- 
ure 5  shows  the  individual  scans  before  averaging, 
and  figure  6  shows  plots  of  the  averaged  signals. 
The  noise  becomes  progressively  higher  with  deeper 
signal  penetration,  as  shown  in  figure  5,  This 
effect  is  due  to  the  internal  noise  of  the  spectrum 
analyzer.    The  signal  to  noise  ratio  of  the  spec- 
trum analyzer  was  constant  for  all  signal  levels, 
since  the  spectrum  analyzer  was  used  in  the  loga- 
rithmic mode.    Thus,  for  higher  signal  levels 


263 


4c 

[l 

1 

1 

0  1  2  3  4  5 

Microseconds 

Fig.  4.    Comparison  of  J-j(t)  calculated  by  eq.  (6) 
(upper  trace)  with  the  multiplied  J-j(t)  de- 
fined by  eq.  (13)  (lower  trace)  for  the 
three  filters,    a)  high  frequency  signals; 
b)  middle  frequency  signals;  c)  low  fre- 
quency signals. 


(longer  gates)  the  absolute  noise  level  increased. 
This  resulted  in  a  decreased  signal  to  noise  ratio 
for  the  differential  power  spectrum  of  higher  level 
signals. 

It  is  interesting  to  note  that  the  time  resolu- 
tion of  the  multiplied  signal  J-j(t)  is  better  than 
the  resolution  of  the  broadband  signal  g(t)  and 
significantly  better  than  the  resolution  of  the 
filtered  signals,  S-j(t)  used  in  the  original  SCU 
implementation  (see  fig.  2).    This  is  interpreted 
as  a  result  of  having  increased  the  information 
bandwidth  of  the  broadband  signal  g(t)  by  the  band- 
width of  the  filtered  waveform  Si(t)  in  the  multi- 
plied signal  Ji(t). 

We  can  also  observe  that  agreement  between  the 
calculated  Ji(t)  and  the  analog  multiplied  Ji(t) 
progressively  improves  as  we  go  from  the  high 
frequency  band  to  the  low  frequency  band.  This 
trend  is  probably  due  to  the  fact  that  the  assump- 
tion of  a  real  filter  function  is  better  justified 
for  the  low  frequency  filter  than  for  the  high 
frequency  filter.    The  time  delay  error  which  was 
approximately  constant  for  the  three  filters,  is 
most  disruptive  to  the  high  bandpass  filter,  be- 
cause of  the  shorter  repeat  period  of  the  filtered 
si  gnal . 

The  results  of  this  experiment  show  only  fair 
agreement  between  the  band  limited  instantaneous 
power  calculated  by  eq.  (6),  and  the  band  limited 
instantaneous  power  produced  according  to  eq.  (13). 
This  result  demands  that  we  reexamine  the  assump- 
tions made  in  the  theoretical  analysis.    The  first 
assumption  of  a  real  valued  filter  function  is 
reasonably  valid  because  of  the  broadband  delay 
line  compensation  for  the  group  delay  through  the 
filters.    The  appearance  of  predominately  positive 
signals  in  the  multiplied  waveforms  indicates  that 
the  filtered  and  broadband  waveforms  are  largely 
in  phase,  though  the  filters  have  some  dispersive 
effects.    The  second  assumption  was  that  the  filter 
is  causal  in  the  time  domain.    This  assumption  also 
appears  valid  for  passive  filters.    However,  close 
examination  reveals  that  the  two  assumptions  above 
are  mutually  exclusive.    That  is,  if  a  filter  func- 
tion is  real  valued  in  the  frequency  domain,  it 
must  be  noncausal  in  the  time  domain.    This  is  true 


2  3 
Microseconds 


2  3 
Microseconds 


Fig.  5.    Individual  calculator  plots  of  J-j(t)  for 
40  scans  through  the  sponge,    a)  high  fre- 
quency band;  b)  middle  frequency  band; 
c)  low  frequency  band. 


Fig.  6.    Calculator  plot  of  the  average  J^-(t)  for 
the  40  scans  in  figure  5.    a)  high  fre- 
quency band;  b)  middle  frequency  band; 
c)  low  frequency  band. 


264 


because  a  real  valued  function,  when  transformed  by 
eq.  (4),  must  have  a  real  part  (the  integral  of  the 
cosine  terms)  that  is  an  even  function  of  time.  An 
even  function  of  time  must  be  noncausal.  Hence, 
the  transition  from  eq.  (12)  to  eq.  (13)  is  not 
val id. 

5.    Time  Reversal  Symmetry 

From  the  above  discussion,  it  appears  impossible 
to  display  Page's  instantaneous  power  spectrum  by 
simple  signal  filtering  and  broadband  multiplica- 
tion.   The  restriction  to  non-causal  filters  re- 
quired by  eq.  (8)  introduces  a  contribution  to  the 
instantaneous  power  from  the  signal  at  future 
times.    The  time  reversal  symmetry  of  the  filter 
function  in  the  time  domain  means  that  eq.  (13) 
corresponds  to  an  average  of  Page's  instantaneous 
power  and  a  similar  instantaneous  power  with  the 
time  scale  reversed  (i.e.,  a  signal  taken  from 
to  time  t).    An  instantaneous  power  spectrum  of 
this  type  has  been  introduced  by  Morris  Levin  [5]. 

Levin  defines  an  instantaneous  power  spectrum 
analogous  to  Page's  except  with  the  added  proper- 
ty of  time  reversal  symmetry.    According  to  Levin, 
the  symmetric  iriStantaneous  power  spectrum  is 
given  by 


where 


1 

2  3t 


2^(03)12  -  |G^(a)) 


G^(a)) 


x)e-J"^dx 


(14) 


(15) 


Ji(t)  =  f  g(t)Re 


C^.  (a))du 


(20) 


Again,  assuming  C^{u)  is  real  valued  rather  than 
complex,  eq.  (20)  becomes 


J.(t)  =  g(t)Re 


y  G(,o)C.(aOej'"^da) 


(21) 


The  integral  in  eq.  (21)  is  the  inverse  Fourier 
transform  of  the  product  G(o))Ci (u) ,  and  hence  be- 
comes the  convolution  of  the  two  corresponding 
time  domain  signals 


J.(t)  =  2^g(t)Re[g(t)*c.(t)j 


or 


J.(t)  =  2ug(t)S.(t) 


(22) 


(23) 


where  S^-(t)  is  defined  by  eq.  (5).    No  assumption 
of  causality  of  c-j(t)  is  required  in  the  derivation 
of  eq.  (23)  because  of  the  time  reversal  symmetry 
of  Levin's  instantaneous  power  spectrum. 

Thus  the  two  modifications  of  the  SCU  system 
stated  above  (broadband  delay  and  signal  multipli- 
cation) produce  a  band  limited  instantaneous  power 
corresponding  to  Levin's  instantaneous  power  spec- 
trum, rather  than  Page's  spectrum  as  initially 
intended. 

6.  Conclusions 


and 


g:J(co)  =  /  g(x)e-j"^dx   .  (16) 
t 

Levin  also  shows  that  p^{t,ui)  is  given  by 

Ps(t,a>)  =  g(t)Re  [6(^)6^"^*]  (17) 


where 

00 

G(w)  =   f  g(t)e"J'"tdt    .  (18) 


G(u)  is  simply  the  frequency  spectrum  of  the  com- 
plete waveform  g(t).    By  analogy  to  eq.  (6)  above, 
we  redefine  the  band  limited  instantaneous  power 
Ji(t)  (for  Levin's  spectrum)  to  be 


The  analysis  we  have  presented  indicates  that 
spectra-color  ultrasonography  should  be  modified 
in  two  ways.    First,  the  filters  used  must  have 
real  filter  functions  in  the  frequency  domain,  in- 
cluding compensation  for  the  time  delays  through 
the  filter.    Second,  the  appropriate  signals  to 
color  code  and  display  are  the  products  of  the 
filtered  signals  with  the  broadband  echo  waveform, 
rather  than  the  simple  filtered  waveforms. 

With  these  two  modifications  in  effect,  SCU  rep- 
resents a  true  low  resolution  instantaneous  spectral 
analysis  of  ultrasonic  echoes  corresponding  to  the 
instantaneous  power  spectrum  of  Morris  Levin.  In 
addition,  the  time  resolution  of  the  new  system 
exceeds  that  of  the  broadband  signal. 

Acknowledgments 

This  work  was  supported  in  part  by  National  In- 
stitute of  Health  Grant  #EY  00224-15  and  The  Ohio 
Lions  Research  Foundation. 

References 


J^it)  =     /   P5(t,a))C.(a))dw  (19) 


where  C^(u))  is  defined  as  before.  Substituting 
eq.  (17)  into  eq.  (19)  gives 


[1]    Holasek,  E.,  Gans,  L.  A.,  Purnell,  E.  W.,  and 
Sokollu,  A.,  A  method  for  spectra-color  B-scan 
ultrasonography,  J.  Clinical  Ultrasound,  3^, 
175-178  (1975). 

[2]    Purnell,  E.  W.,  Sokollu,  A.,  Holasek,  E.,  and 
Cappaert,  W.  E.,  Clinical  spectra-color  ultra- 
sonography, J.  Clinical  Ultrasound,  3^,  187-189 
(1975). 


265 


[3]    Page,  C.  H. ,  Instantaneous  power  spectra,  J_. 
Applied  Physics,  23,  103-106  (1952). 

[4]    Holasek,  E.,  Jennings,  W.  D.,  Sokollu,  A.,  and 
Purnell ,  E.  W.,  Recognition  of  Tissue  Patterns 
by  Ultrasonic  Spectroscopy,  Ultrasonic  Sym- 
posium Proceedings, 'IEEE  Cat.  #73  CH  0807-8SU, 
73-76  (1973). 

[51    Levin,  M.  J.,  Instantaneous  Spectra  and  Ambi- 
guity Functions,  IEEE  Transactions  Information 
Theory,  Vol.  IT-10,  95-97,  (1964). 


266 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C. ,  1979). 


IDENTIFICATION  OF  TISSUE  PARAMETERS  BY  DIGITAL  PROCESSING  OF 
REAL-TIME  ULTRASONIC  CLINICAL  CARDIAC  DATA 


L.  Joynt,  D.  Boyle,  H.  Rakowski,  R.  Popp,  and  W.  Beaver 

Center  for  Integrated  Electronics  in  Medicine 
and  Division  of  Cardiology 
Stanford  University 
Stanford,  California    94305,  U.S.A. 


A  study  to  assess  the  feasibility  of  obtaining  diagnostical ly  useful  tissue  charac- 
terization information  by  digitally  processing  clinical  cardiac  data  is  described. 
Normal  subjects,  myocardial  infarction,  IHSS,  and  amyloid  patients  were  studied.  The 
data  acquisition  system  used  to  record  data  from  a  real-time  scanner  is  also  described. 
Significant  changes  in  the  RF  signals  and  frequency  spectra  as  the  heart  moves  were 
noted  over  very  short  time  intervals,  indicating  the  need  for  a  dynamic  tissue  charac- 
terization measure.    Wide  variation  in  the  spectral  characteristics  of  the  signals  from 
the  normal  population  were  found.    Behavior  of  the  spectra  for  the  MI  patient  data  were 
noted  which  differentiated  them  from  the  other  subjects. 


Key  words:    CI inical  cardiac  data;  Fast  Fourier  Transform;  digital  processing;  frequency 
spectra;  in  vivo;  microprocessor-controlled  data  acquisition;  myocardial  in- 
farction; real-time;  tissue  characterization;  ultrasound  diagnosis. 


1.  Introduction 

The  possibility  of  obtaining  diagnostic  informa- 
tion by  processing  clinical  ultrasound  data  from 
cardiac  patients  is  explored  in  this  paper.  Lele 
et  al.  [l]i  and  Yuhas  et  al.  [2]  have  differentiat- 
ed between  normal  and  ischemic  or  infarcted  heart 
muscle  in  vitro  on  the  basis  of  measurements  of 
acoustic  attenuation.    Extension  of  these  methods 
to  clinical  ultrasound  examinations  is  complicated 
by  factors  including  cardiac  motion,  intervening 
tissue,  and  the  angle  of  incidence  of  the  ultra- 
sound beam  to  the  interrogated  tissue  [3].  The 
most  direct  approach  to  exploring  the  effects  of 
these  difficulties  in  the  clinical  environment  is 
to  interface  to  an  instrument  in  clinical  use. 
Coupling  a  real-time  scanner  to  a  high-speed  data 
acquisition  system  makes  it  possible  to  obtain  in- 
formation on  the  dynamic  state  of  the  heart.  A 
preliminary  study  on  a  limited  number  of  patients 
and  normal  subjects  was  done  in  order  to  assess  the 
feasibility  of  deriving  diagnostically  useful  in- 
formation about  the  type  and  state  of  cardiac  tis- 
sue by  digital  processing  of  clinically  acquired 
data. 

2.    Data  Acquisition 

A.    The  Data  Acquisition  System 

The  design  of  the  data  collection  system  for 
this  project  is  based  on  a  number  of  considerations. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


In  order  to  eliminate  possible  information  loss  due 
to  the  detection  process,  the  rf  signal  is  digitized 
for  off-line  computer  processing.    Experience  with 
an  earlier  version  of  the  data  system  had  shown  that 
large  changes  occur  in  signals  recorded  at  100  ms 
intervals.    Accordingly,  the  current  version  was 
designed  to  record  as  continuously  as  feasible  so 
that  these  changes  could  be  observed  in  more  detail. 
Thus,  large  quantities  of  data  must  be  recorded  at 
a  very  high  rate.    A  compromise  solution,  trading 
off  the  speed  of  semiconductor  memory  for  the  low 
cost  of  tape  memory  by  using  both  high  and  low 
speed  semiconductor  memory  to  buffer  the  input  to  a 
magnetic  tape  drive,  is  used  in  the  present  system. 
A  Biomation  8100  Transient  Recorder  is  used  as  the 
A/D  converter  and  high  speed  buffer.    The  data  is 
then  transferred  to  a  large  slow  speed  buffer,  con- 
sisting of  32  kilobytes  of  microcomputer  memory, 
which  stores  data  from  a  number  of  scan  lines  for 
subsequent  recording  on  a  magnetic  tape  drive  having 
a  transfer  rate  of  30  kilobytes/second.    This  device 
provides  mass  storage  and  convenient  input  to  a  com- 
puter for  later  processing. 

The  data  acquisition  performance  of  the  system 
is  determined  by  the  limitations  of  the  digital  hard- 
ware and  by  the  characteristics  of  the  scanner. 
When  the  Varian  real-time  sector  scanner  is  used  as 
the  signal  source,  data  may  be  acquired  from  any  one 
cursor-selected  scan  line  every  600  microseconds  or 
from  the  sequence  of  scan  lines  comprising  a  frame 
every  30  milliseconds.    The  rf  signal  is  digitized 
at  20  MHz  so  that  about  100  microseconds  or  7  cm  of 
data  can  be  stored  in  the  transient  recorder's  2048 
word  memory.    Only  those  samples  actually  correspond- 
ing to  the  area  of  interest  are  transferred  to  the 
microcomputer  memory  buffer.    The  total  time  to  re- 


267 


cord  data  from  one  scan  line  is  under  2  milliseconds. 
Thus,  a  substantial  portion  of  a  frame  can  be  digi- 
tized at  real-time  rates.    The  system  also  includes 
the  option  to  record  a  sequence  of  lines  at  30  ms 
intervals  in  order  to  cover  a  complete  cardiac  cycle 
in  one  buffer.    When  the  data  buffer  is  filled  the 
recording  process  is  interrupted  for  a  few  seconds 
while  the  data  is  transferred  to  magnetic  tape. 

Data  acquisition  is  triggered  at  a  pre-selected 
point  of  the  cardiac  cycle.    A  simultaneous  Polaroid 
picture  can  be  obtained.    If  more  than  one  buffer  is 
required,  subsequent  buffers  also  begin  at  the  select- 
ed point  of  the  cardiac  cycle. 


CRT 
TERMINAL 


ULTRASONIC 
SCANNER 


MICRO-COMPUTER 


TRANSFER  AT  30  KB/s 


INTERFACE 
AND 

DMA  CONTROLLER 


9-TRACK  MAGNETIC 
TAPE 


TRANSFER  AT  2  MB/s 


SAMPLED  AT  20  MB/s 


BIOMATION  8100 
TRANSIENT  RECORDER 


MAGUS  SYSTEM 

-  CROSS-ASSEMBLER 

-  SIGNAL  PROCESSING 

-  GRAPHIC  OUTPUT 


Fig.  1.    Block  diagram  of  the  data  acquisition 

system.    The  RF  signal  is  sampled  at  20 
MHz,  filling  up  the  transient  recorder 
buffer  at  20  Megabytes/second,  trans- 
ferred to  microcomputer  memory  at  2 
Megabytes/second,  and  transferred  to 
magnetic  tape  at  30  kilobytes/second. 

The  data  system  has  been  designed  to  be  easy  to 
use  in  a  clinical  environment.    The  operator  can 
interact  with  the  microcomputer  software  system  to 
specify  the  data-taking  procedure.  Information 
about  the  portion  of  the  frame  to  be  digitized  is 
stored  in  a  table  and  written  on  the  tape  along  with 
the  data  for  use  in  processing.    Comments  can  be  ad- 
ded from  the  operator's  CRT  terminal  to  record  ad- 
ditional information  about  the  tissue  under  study. 

B.    Data  Acquisition  Procedure 

A  variety  of  patients  with  different  myocardial 
pathologies  were  chosen  to  exhibit  various  degrees 
of  focal  or  diffuse  replacement  of  normal  myocardium. 
Included  were  patients  with  transmural  myocardial 
infarction  (2),  idiopathic  hypertrophic  subaortic 
stenosis  (IHSS)  (3),  and  cardiac  amyloidosis  (1). 
Three  normal  subjects  were  studied  as  controls. 

Consideration  of  changes  in  the  echoes  due  to 
motion  and  muscle  contraction  during  the  cardiac 
cycle  is  particularly  important  in  a  clinical  study 
of  cardiac  patients.    In  order  to  study  these  changes 
single  scan  lines  through  the  septum  and  posterior 
wall  of  the  left  ventricle  were  recorded  at  2  ms  and 


30  ms  intervals.    Data-taking  was  synchronized  with 
the  patient's  electrocardiogram  to  commence  at  end 
systole  and  end  diastole.    In  the  patients  with 
myocardial  infarction,  the  area  of  infarction  was 
identified  as  an  akinetic  left  ventricular  segment 
corresponding  to  the  region  denoted  by  Q  waves  on 
the  electrocardiogram.    Data  was  collected  from  this 
segment  and  compared  to  normally  contracting  seg- 
ments in  the  same  patient.    In  the  patients  with 
IHSS,  the  region  of  asymmetric  septal  hypertrophy 
could  easily  be  identified  and  often  has  an  abnormal 
ground-glass  appearance  when  viewed  in  the  real-time 
image  [4].    Again,  comparison  was  made  with  adjacent 
areas  of  normal  myocardial  appearance  and  thickness 
and  with  the  posterior  left  ventricular  wall.  In 
the  patient  with  diffuse  myocardial  involvement  with 
cardiac  amyloidosis,  representative  areas  of  inter- 
ventricular septum  and  posterior  left  ventricular 
free  wall  were  studied. 

3.    Data  Analysis 

The  recorded  ultrasonic  data  was  processed  on  an 
HP21MX  minicomputer.    Fourier  transforms  of  each  re- 
corded data  trace  were  computed  using  the  Fast 
Fourier  Transform  (FFT)  algorithm.    By  taking  the 
Fourier  transform  of  a  single  echo,  one  obtains  in- 
formation about  the  overlying  tissue  through  which 
the  acoustic  pulse  has  passed,  while  the  Fourier 
transform  over  an  interval  containing  several  echoes 
gives  information  about  the  spatial  structure  of  the 
tissue  through  interference  effects.  Periodicities 
of  the  spectral  peaks  due  to  constructive  inter- 
ference between  scatterers  spaced  by  a  distance  d 
will  be  spaced  by  Af  =  c/2d,  where  c  is  the  acoustic 
velocity. 

The  system  impulse  response  of  the  Varian  scanner 
is  illustrated  in  figure  2.    The  limited  bandwidth, 
from  about  1.8  to  2.9  MHz,  has  implications  for 
both  kinds  of  tissue  information  sought.    In  vitro 
studies  by  other  workers  [1-8]  have  shown  that  dif- 
ferences in  frequency-dependent  attenuation  between 
normal  and  ischemic  or  infarcted  heart  tissue  are 
small  over  this  range.    Only  periodicities  cor- 
responding to  spatial  separations  larger  than  .75 
mm  will  appear  in  this  narrow  bandwidth.    Thus  the 
available  tissue  characterization  information  would 
be  derived  from  observing  shifts  in  the  overall 
shape  of  the  spectrum. 

A  primary  interest  was  to  note  the  effects  of 
cardiac  motion  on  the  spectra  of  returned  echoes. 
We  tried  to  assess  changes  in  the  spectra  as  a 
function  of  time  in  order  to  see  possible  differ- 
ences in  the  spectra  corresponding  to  the  anatomic 
differences  between  the  normal  and  diseased  heart 
muscle.    Figures  3  through  7  show  examples  of  the 
data  processing  sequence  for  the  long  axis  view  of 
the  septum  at  end  diastole  for  one  patient  in  each 
of  the  groups  studied. 

4.  Results 

Substantial  changes  in  the  RF  signals  returned 
from  the  heart  and  their  spectra  were  observed  over  > 
time  intervals  short  compared  to  the  cardiac  cycle. 
Data  recorded  at  2  ms  intervals  appeared  to  ade- 
quately represent  the  dynamically  varying  signal, 
whereas  data  recorded  at  30  ms  changed  abruptly  and 
apparently  randomly  between  traces.    Figures  8  and 
9  demonstrate  this  effect.    A  dynamic  characteriza- 
tion of  heart  tissue  is  probably  indicated  since 
parameter  measurements  derived  from  the  frequency 


268 


02468         10        0  1  2  3  4  5 

MICROSECONDS  FREQUENCY  (MHz) 

Fig.  2.    a)  Impulse  response  of  Varian  Real-time  Sector  Scanner  measured  by  recording  a  single 
echo  returned  from  the  front  side  of  a  Lucite  block  in  a  water  tank.    This  echo  was 
digitized  at  100  MHz.    b)  1024  point  FFT  of  impulse  response. 


MICROSECONDS  FREQUENCY  (MHz) 

Fig.  3.    a)  Photograph  of  the  scanner  image.    The  rf  signal  was  recorded  along  the  scan  line 

indicated  by  the  bright  cursor  line  on  the  image,    b)  The  recorded  rf  trace  is  synchronized 
to  the  desired  point  in  ECG  and  to  the  picture  of  the  scanner  image.    The  operator  chooses 
the  origin  of  the  time  scale.    Here,  echoes  from  the  septum  were  recorded,    c)  512  point  FFT. 
The  portion  of  the  rf  trace  used  in  the  FFT  is  outlined  in  figure  b. 


spectrum  at  one  instant  of  time  may  not  adequately 
describe  the  same  tissue  30  milliseconds  later. 

The  first,  second,  and  third  central  moments 
were  used  to  characterize  the  spectrum  of  each  tis- 
sue sample,  permitting  study  of  changes  in  the  shape 
of  the  spectrum.    Differential  frequency-dependent 
attenuation,  as  indicated  by  a  shift  in  the  first 
moment,  for  normal  versus  damaged  myocardium  was  not 
observed.    In  fact,  attenuation/centimeter  appeared 
to  be  constant  across  the  frequency  range  involved. 
It  is  likely  that  frequency  dependent  effects  are 
overshadowed  by  the  effects  of  imaging  a  tissue 
volume  deep  within  the  body,  due  to  overlying  tis- 
.sue,  diffraction  spreading  of  the  beam,  and  orienta- 


tion of  the  tissue  to  the  ultrasound  beam.  There 
was  a  wide  range  of  normal  spectral  moments  which 
overlapped  those  of  the  diseased  patients.  Disap- 
pointingly, the  IHSS  tissue,  which  presents  a  dif- 
ferent appearance  on  the  scanner  image,  was  not 
distinguishable  from  the  normal  by  the  shape  of  the 
Fourier  spectra.    A  comparison  of  normal  and  IHSS 
tissue  spectra  is  depicted  in  figure  10.    The  myo- 
cardial infarction  (MI)  patients  were  set  apart 
from  the  others  by  much  less  fluctuation  in  their 
spectral  mean  and  variance  curves  (fig.  11).  This 
implies  that  the  tissue  is  moving  or  contracting 
less  than  normal  tissue  which  correlates  well  with 
the  observed  image  on  the  real-time  display.  There 


269 


(c) 


0  10  20  30  40       0         1  2  3         4  5 

MICROSECONDS  FREQUENCY  (MHz) 

Fig.  4.    a)  Photograph  of  scanner  image,  b)  Recorded  rf  signal  from  selected  scan  line,  c)  512  point  FFT 


AMYLOID 


(b) 


(c) 


Mil*- 


10  20  30 

MICROSECONDS 


40      0         1  2  3  4 

FREQUENCY  (MHz) 


Fig.  5.    a)  Photograph  of  scanner  image,  b)  Recorded  rf  signal  from  selected  scan  line,  c)  512  point  FFT 


(a) 

NORMAL  A 


10  20  30 

MICROSECONDS 


40  0 


12  3  4 

FREQUENCY  (MHz) 


Fig.  6.    a)  Photograph  of  scanner  image,  b)  Recorded  rf  signal  from  selected  scan  line,  c)  512  point  FFT 


270 


(a1 
NORMAL 


(b) 


0 


10 


20  30 
MICROSECONDS 


40 


Fig.  7. 


2  3 
FREQUENCY  (MHz) 

a)  Photograph  of  sccinner  image,  b)  Recorded  rf  signal  from  selected  scan  line,  c)  512  point  FFT. 


Fig.  8. 


10         20  30 

MICROSECONDS  

0  12  3 

FREQUENCY  (MHz) 

a)  Sequence  of  rf  echo  traces  from  normal  septum.  Signals  were  recorded 
at  2  ms  intervals.  Bottom  trace  occurred  first  in  time,  b)  FFT  spectra 
of  signals  in  a. 


was  also  a  spectral  shift  of  approximately  100  kHz 
in  the  means  of  the  spectra  between  the  long  axis 
and  short  axis  views  for  the  MI  data.    This  is  like- 
ly due  to  a  more  tangential  view  of  the  tissue  in 
short  axis,  demonstrating  the  substantial  effect  of 
orientation  to  the  ultrasound  beam. 

5.  Conclusions 

Straightforward  application  of  methods  [1,2,8] 
which  have  differentiated  between  normal  and  ische- 
mic or  infarcted  myocardium  in  vitro  to  in  vivo 


data  derived  from  current  clinical  instruments  is 
not  likely  to  be  successful  due  to  the  triple 
whammy  of  limited  bandwidth,  overlying  tissue,  and 
cardiac  motion.    Spectrum  skewing  due  to  broad  fre- 
quency interference  effects  and  analysis  of  the  time 
sequence  of  spectral  changes  may  show  some  utility. 
The  substantial  changes  in  the  rf  and  frequency 
spectra  over  very  short  time  intervals  lead  to 
speculation  that  they  reflect  the  state  of  contrac- 
tion of  cardiac  muscle  as  well  as  changes  in  geome- 
try.   The  data  acquisition  system  described  in  this 
paper  represents  a  significant  advance  in  the  state 


271 


2.32  MHz 


2.30  MHz 


2.25  MHz 


Fig.  9.    a)  Four  frequency  components  of  the  FFT  of  normal  septum  data  recorded  at  2  ms  intervals. 

Each  component  is  plotted  as  a  function  of  time,  b)  Same  four  frequency  components  plotted 
as  a  function  of  time  for  data  recorded  at  30  ms  intervals. 


2.60- 


2.40 


2.20 


MEAN 


VARIANCE 


42 

MILLISECONDS 

Fig.  10.    Comparison  of  mean  and  variance  of  spectra  of  normal  and  IHSS  spetal  tissue. 
Moments  are  plotted  as  a  function,  2  ms  per  point. 


of  the  art  of  the  digital  recording  of  ultrasound 
rf  in  terms  of  flexibility  and  ease  of  use  as  well 
as  in  the  kind  of  data  it  makes  available. 

Acknowledgment 

This  work  was  supported  by  the  U.S.  Department 
of  Health,  Education,  and  Welfare  under  grant  GM- 
17940. 


References 

[1]    Lele,  P.  P.,  Mansfield,  A.  B.,  Murphy,  A.  I., 
Namery,  J.,  and  Senapti ,  N.,  Tissue  Charac- 
terization by  Ultrasonic  Frequency-Dependent 
Attenuation  and  Scattering,  in  Ultrasonic  Tis- 
sue Characterization,  M.  Linzer,  ed..  National 
Bureau  of  Standards  Spec.  Publ .  453,  pp.  167- 
196  (U.S.  Government  Printing  Office,  Washing- 


272 


LONG  AXIS 

MYOCARDIAL  INFARCTION  A 
SHORT  AXIS 


28  42 
MILLISECONDS 


LONG  AXIS 


MYOCARDIAL  INFARCTION  B 


SHORT  AXIS 


Fig.  11.    Means  of  the  spectra  of  data  from  two  infarcted  spetums.    Data  is  plotted 
at  2  millisecond/point  for  the  long  and  short  axis  views  for  both  patients. 


ton,  D.C.,  1976).  [6] 

Yuhas,  D.  E.,  Mimb,  J.  N.,  Miller,  J.  G.,  Wiess, 
A.  N. ,  and  Sobel ,  B.  E.,  Changes  in  Ultrasonic 
Attenuation  Indicative  of  Regional  Myocardial  [7] 
Infarction  in  Ultrasound  in  Medicine,  D.  White, 
ed. ,  Vol.  3  (Plenum  Press,  New  York,  1977). 

[8] 

Chivers,  R.  C.  and  Hill,  C.  R. ,  A  spectral  ap- 
proach to  ultrasonic  scattering  from  human  tis- 
sue:   methods,  objective  and  backscatteri ng 
measurements,  Phys.  Med.  Biol .  20  (5),  799-815 
(1975).  [9] 

Rossen,  R.  M. ,  Goodman,  D.  J.,  Ingam,  R.  E., 
and  Popp,  R.  L.,  Echocardiographi c  criteria  in 
the  diagnosis  of  idiopathic  hypertrophic  sub- 
aortic stenosis,  Circulation  50,  747-751  (Oct. 
1974).  [10] 

Cooley,  J.  W.  and  Tukey,  J.  W.,  An  algorithm 
for  the  machine  calculation  of  complex  Fourier 
series.  Math.  Computation  19,  297-301  (1965). 


Oppenheim,  A.  V.  and  Schaefer,  R.  V.,  Digital 
Signal  Processing  (Prentice-Hall,  Inc.,  Engle- 
wood  Cliffs,  N.  J.,  1975). 

Crawford,  F.,  Waves  (McGraw-Hill,  New  York, 
1968) . 

Namery,  J.  and  Lele,  P.  P.,  Ultrasonic  Detec- 
tion of  Myocardial  Infarction  in  Dog,  in 
Proceedings  of  1974  Ultrasonic  Symposium,  p. 
491  (IEEE  Cat.  No.  72  CHO  7088  SU,  1974). 

Mohammed,  A.  and  Smith,  R.  G.,  Data  Windowing 
in  Spectral  Analysis,  DREA  Report  75/2,  De- 
fense Research  Establishment  Atlantic,  Dart- 
mouth, N.  S.,  Research  and  Development  Branch, 
Department  of  National  Defense,  Canada. 

Fields,  S.  and  Dunn,  F.,  Correlation  of  echo- 
graphic  visualizability  of  tissue  with  bio- 
logical composition  and  physiological  state, 
J.  Acoust.  Soc.  Am.  54,  809-811  (1973). 


273 


i' 
f 

fl 
I 
p 
il 

pi 
fi 
ii 

tt 
fl 

Ii 
ii 

SI 

II 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


DYNAMIC  AUTOCORRELATION  ANALYSIS  OF  A-SCANS  IN  VIVO 


J.  C.  Gore,^  S.  Leeman,^  C.  Metreweli,^  N.  J.  Plessner,^ 
and  K.  Willsoni 


^Department  of  Medical  Physics 
2 Department  of  Diagnostic  Radiology 

Royal  Postgraduate  Medical  School 
Hammersmith  Hospital,  London,  England 


A  realistic  tissue  model  has  been  analysed  to  indicate  the  information  that  may  be 
derived  from  autocorrelation  studies  of  in  vivo  A-scan  echograms,  as  well  as  to  show 
some  of  the  limitations  of  such  techniques.    Although  many  of  these  difficulties  are 
not  easily  overcome  when  considering  single,  or  even  averaged,  realisations  of  the 
autocorrelation  function  (ACF),  an  analysis  of  the  time  course  of  the  ACF  is  less 
subject  to  such  objections. 

Several  potentially  useful  clinical  applications  of  such  temporal  changes  are  being 
investigated.    These  include  measurements  of  echoes  from  within  heart  muscle  through- 
out the  cardiac  cycle  as  a  possible  indicator  of  cardiac  disease;  similar  variations 
in  echoes  from  the  stomach  wall  during  gastric  emptying  are  demonstrated,  and  a  relation- 
ship to  contractile  state  is  postulated. 

Results  are  also  presented  which  indicate  that  the  perfusion  of  tissues  with  blood 
may  be  assessed  by  this  technique,  and,  on  a  different  physiological  time  scale,  changes 
in  the  placenta  throughout  pregnancy  have  been  investigated.    The  technique  is  being 
extended  to  the  study  of  the  response  of  malignant  tumours  to  treatment. 


Key  words:    A-scan;  correlation  analysis;  temporal  changes. 


1.  Introduction 

Tissue  characterisation  may  be  defined  as  the 
attempt  to  establish  what  quantitative  parameters, 
other  than  range  information,  may  be  extracted 
from  ultrasound  probing  of  human  tissue;  the 
quantities  derived  should  reflect  intrinsic  tissue 
properties  and  in  particular  its  physiological 
state,  be  useful  diagnostically,  or  for  pre-  and 
post-operative  assessment,  or  for  monitoring 
response  to  treatment.    The  attempt  is  not  neces- 
sarily restricted  to  an  ultrasound  technique,  but 
this  modality  is  particularly  attractive,  not  only 
for  the  usual  reasons,  but  also  because  it  pro- 
vides the  facility  for  receiving  and  analysing  a 
signal  from  a  reasonably  localised  region  of  tis- 
sue, as  well  as  providing  images  to  enable  that 
region  to  be  accurately  placed.    Clearly,  only 
physical  properties  of  tissue,  or  their  possible 
change  with  time,  are  amenable  to  detection--often 
micro-structural  features  are  of  interest,  but 

1    other  possibilites  (e.g. ,  temperature  measurement) 
are  certainly  not  excluded. 

Isolated  echo  structure  may  be  probed  in  tissue 

'     characterisation  analyses  [l-3]3,  even  though  the 


extracting  of  purely  physical,  as  opposed  to  geo- 
metrical, data  is  more  complicated  than  may  appear 
at  first  sight  [3].    Here  we  will  restrict  our- 
selves, as  have  many  others,  to  the  scattered  low- 
level  signals  from  a  region  of  tissue,  and  perform 
what  may  be  termed  tissue  characterisation  by 
"echo  ensemble"  analysis.    The  extent  to  which 
these  two  methods  are  complementary  is  not  known 
at  present,  but  it  may  be  noted  that  both  are  im- 
plicit in  qualitative  attempts  to  extract  diagnos- 
tically useful  data  from  grey-scale  images. 

2.    Some  Theoretical  Ideas 

A  conceptual  basis  for  the  empirical  approach 
may  be  laid  from  theoretical  considerations,  just 
as  the  impedance  mismatch  concept  simplifies  the 
interpretation  of  much  of  conventional  ultrasound 
scanning. 

The  tissue  model  adopted  here  comprises  an  in- 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


275 


homogeneous  medium,  with  local  density  and  com- 
pressibility fluctuations  generating  the  ultra- 
sound scattering.    In  this  context,  the  aim  of  tis- 
sue characterisation  by  echo  ensemble  analysis  is 
to  uncover  the  three-dimensional  spatial,  and  even 
temporal,  distribution  of  these  fluctuations;  while 
the  hope  of  the  technique  is  that  all,  or  preferab- 
ly some,  of  the  information  is  sufficient  to  estab- 
lish unambiguously  the  state  of  the  tissue. 

A  convenient  starting  point  is  the  wave  equa- 
tion for  ultrasound  propagation  in  a  medium  with 
stationary  inhomogeneities  of  the  type  described 
above  [4]. 

^^P-E^0=H^T(r)0+V,[p(r)Vp]  (1) 

where  p{r_,t)  is  the  acoustic  pressure  at  location 
£  and  time  t; 
v{r)    is  the  density  fluctuation, 

y  =  (p  -  Pq)/p 

with  po  the  mean,  and  p(r)  the  exact  local  density; 
y(r)  is  the  compressibility  fluctuation, 

Y   =    (k   -  K^)/K^ 

with  Kp  the  mean,  and  K{rJ  the  exact  local  com- 
press ibi  1  ity. 

Cq  is  the  mean  velocity, 

'I  -  (Vo)-^  • 

The  right-hand  side  of  eq.  (1)  is  conventional- 
ly called  the  source  term,  since  it  is  easily  de- 
monstrated [5]  to  be  a  "source"  of  waves,  and  it 
describes  the  interaction  of  the  medium  with  the 
ultrasound  field.    The  source  term  depends  on  both 
the  fluctuations  in  the  medium  and  the  local,  in- 
stantaneous value  of  the  pressure  wave,  as  it  must 
do  if  it  is  to  describe  a  scattering  situation. 
The  tissue  model  described  by  eq.  (1)  may  be  shown 
[5],  to  contain  that  of  Waag  and  Lerner  [6]  as  a 
special  case.    Since  the  scattering  is  known  to  be 
weak,  both  y  and  y  may  be  taken  to  be  small  quan- 
tities.   Any  scattering  calculation  requires  the 
initial  conditions  to  be  clearly  specified,  and 
we  have  adopted  an  initial  pul  se  pin  i  ncident 
along  the  Z-direction,  of  the  form: 

p^^(r,t)  =  a(z  -  CQt)b(h)exp|i(k^z  -  co^t)}  (2) 

where  a  is  the  (axial)  pulse  shape,  b  is  the  beam 
profile,  h  is  the  spatial  coordinate  transverse  to 
Z,  ko  is  the  carrier  wave  vector  amplitude,  and 
'"0  ~  Coko-    Eq.  (2)  is  a  good  approximation  to 
pulses  from  diagnostic  machines  over  much  of  the 
range  of  interest. 

For  weak  scattering,  and  for  weakly  focused  in- 
cident fields,  the  far-field  final  result  [7]  shows 
that  the  back-scattered  pressure  amplitude  is 
generated  by  the  pulse-smoothed  density/compressi- 
bility fluctuations  in  the  medium.    The  power 
spectrum,  of  the  backscattered  echoes  (with  bound- 
ary effects  neglected),  is  given  by: 

|p^(.)|....||A(^.ko)p  |r(|)p  (3) 


where  A  is  the  Fourier  transform  of  the  axial 
pulse  shape,  u'*  represents  a  "Rayleigh"  factor, 
and  the  effect  of  the  medium  is  contained  only  in 
the  structure  factor. 


oo  211 

<V>  =  f      f  dehb(h)y(h,e,z) 

0  0 

and  similarly  for  r  .    6  is  the  angle  variable  in 
the  cylindrical  polar  coordinates  (h,e,Z)  used  in 
the  calculation.    The  Z-integration  may  be  extend- 
ed to  infinite  limits,  since  the  fluctuations  are 
presumed  to  vanish  outside  the  (finite)  scattering 
volume.    The  structure  factor  r  is  the  spatial 
frequency  spectrum  of  the  beam-profile  smoothed 
tissue  structure,  and  it  is  seen  that,  for  the 
backscattering  case,  the  density  and  compressibili- 
ty fluctuations  contribute  equally,  but  with  dif- 
ferent phase.    At  other  scattering  angles  the  two 
types  of  fluctuations  will  contribute  with  differ- 
ently varying  magnitudes,  and  they  may  be  regard- 
ed as  two  different  types  of  scattering  element: 
the  one  (compressibility)  scattering  as  an  acous- 
tic monopole,  and  the  other  (density)  behaving  as 
an  acoustic  dipole.    This  behaviour  may  be  deduced 
from  the  form  of  the  source  term  in  eq.  (1),  or 
more  directly,  from  straightforward  physical  argu- 
ments [5].    Since  the  relative  importance  of  the 
two  types  of  scattering  element  varies  with  direc- 
tion, different  observers  employing  a  fixed-angle 
scattering  technique  may  obtain  apparently  con- 
flicting results  from  the  same  tissue.    In  general, 
at  least  six  independent  one-dimensional  experi- 
mental functions  would  have  to  be  measured  in 
order  to  reconstruct  both  y  and  u  from  the  data, 
and  since  this  would  seem  to  be  impracticable,  it 
is  emphasized  that,  for  the  characterization  of 
anisotropic  tissues  by  fewer  independent  func- 
tions, it  is  essential  to  specify  internal  tis- 
sue landmarks  with  respect  to  which  scanning 
planes  and  directions  may  be  fixed.    If  not,  in- 
tercomparison  of  results  between  different  work- 
ers, and  the  repeatability  of  a  single  investi- 
gator's findings,  cannot  be  established. 

Quite  apart  from  difficulties  inherent  in 
trying  to  quantitate  the  effect  of  overlying  tis- 
sues and  interfaces,  the  above  calculation  shows 
that  pulse  shape,  beam  profile,  and  carrier  fre- 
quency are  important  quantities  which  fundamental- 
ly influence  the  scattered  pressure  power  spec- 
trum, and  which  must  be  specified  or,  better,  de- 
convoluted  from  the  data.    Moreover,  in  regions 
where  the  incident  pulse  is  sharply  focused 
(|vp^p|  large),  scattering  from  the  density  fluc- 
tuations becomes  more  prominent;  this  may  well 
prescribe  a  suitable  technique  for  mapping  the 
density  and  compressibility  scattering  elements 
individually,  if  so  desired.    In  addition,  the 
measured  backscattering  is  further  modified  by 
the  transfer  properties  of  the  receiver  [8],  but 


276 


all  these  "system"  artifacts  are  measurable,  and 
should  be  prescribed,  if  not  specifically  allowed 
for. 

Pulse  smoothing  implies  a  certain  loss  of  fine 
detail:    scattering  experiments  do  not  indicate 
the  tissue  micro-structure  (<<A)  as  such.    It  is 
quite  clear  from  the  above  that  measured  pressure 
power  spectra  from  the  same  tissue  may  differ  with 
different  transducer  apertures  and  excitations, 
but  it  is  also  worth  emphasizing  that,  in  prin- 
ciple, the  same  backscattered  power  spectrum  may 
be  seen  from  di ff erent  tissues,  with  different 
density  and  compressibility  fluctuations,  provid- 
ed that  Ty-ry  remains  the  same.    This  merely  under- 
lines the  fact  that,  even  for  isotropic  media,  a 
single  backscattering  experiment  is  in  general  in- 
sufficient to  characterize  tissue  unambiguously. 
Fortunately,  in  practice,  the  situation  may  be 
less  complicated  as  there  is  some  indication  that, 
for  most  tissues,  only  the  compressibility  (i.e., 
elasticity)  fluctuations  are  of  importance  for 
the  production  of  echoes  [9]. 


pen  (SAC  Graf  Pen),  interfaced  to  a  HP  2100A  com- 
puter via  a  Tektronix  4010  visual  display  unit. 
With  this  arrangement  effective  sampling  rates  of 
up  to  15  MHz  are  achieved,  although  at  present  it 
is  more  usual  to  sample  echoes  from  a  region 
'V  2  cm  in  extent  at  a  rate  of  Ih  MHz,  consistent 
with  providing  estimates  of  the  ACF  at  lag  inter- 
vals corresponding  to  100  pm  of  tissue. 

The  discrete,  normalised  ACF,  cj,  is  estimated 
from  N  echo  samples.  Pi,  using  the  unbiased  esti- 
mator [10], 


r=—  Vp-P        P  -P 
J     N  1       i-,M     i+j  i+j,M 

where 

1  i+M 
^•,M  =  (2M  +  1)  .^^  Pi 


3.  Method 

As  indicated  in  eq.  (3),  by  measuring  the  power 
spectrum  of  echoes,  the  structure  factor,  r,  may 
in  principle  be  determined  over  a  limited  band  of 
spatial  frequencies  to  within  an  accuracy  set  only 
by  noise  and  the  uncertainty  in  knowledge  of  |A|. 
However,  we  have  chosen  to  explore  the  use  of  the 
time-domain  autocorrelation  function  (ACF)  of 
echo  amplitudes;  as  well  as  being  easier  to  cal- 
culate, the  ACF  contains  all  the  information  in- 
cluded in  the  power  spectrum,  and  although  the 
errors  inherent  in  its  estimation  differ  from 
those  in  power  spectrum  calculations,  it  seems 
likely  that  on-line  ACF  measurement  may  be  at 
least  as  suitable  for  "in  vivo"  tissue  characteri- 
zation in  real  time  as  Fourier  methods.    The  de- 
sired autocorrelation  function  is  given  by 


c(t)  =    /    p(t)p(t  +  T)dt 


where  p(t)  =  the  backscattered  acoustic  pressure 
amplitude  at  time  t. 

From  eq.  (3),  it  can  be  shown  that 

c(t)  -  6(t)*?(t) 

where  *  denotes  convolu^tion,_ 

6(t)  =  the  ACF  of  <y>  -  <y> 

and  is  an  indicator  of  tissue  structure. 

5(t)  =  the  ACF  of  the  second  derivative  of  the 
impulse  response  of  the  pulse-echo  sys- 
tem [8]. 

The  convolution  with  the  system  function,  limits 
the  information  available  unless  steps  are  taken  to 
remove  its  influence,  but  such  further  processing 
has  not  yet  been  attempted  by  us. 

The  equipment  used  in  these  studies  is  a  Nuclear 
Enterprises  NE  4102  Diasonograph ,  operated  with 
weakly  focused  transducers  centred  on  2.5  MHz,  with 
total  system  bandwidth  approximately  W  MHz.  Sig- 
nals from  the  A-scan  receiver  are  at  present  re- 
corded photographically  from  an  oscilloscope,  and 
the  echo  envelope  is  then  digitised  with  a  sonic 


is  the  running  mean  of  the  echo  amplitude.  The 
moving  average  is  calculated  over  a  distance  that 
is  sufficiently  large  not  to  influence  the  fine 
structure  of  the  ACF  estimate  but  which  filters 
spurious  low  frequency  fluctuations  and  strengthens 
stationari ty.    Such  detrending  is  performed  typical- 
ly over  intervals  of  ■v  1  cm. 

4.    Data  Analysis 

In  general,  the  entire  ACF  must  be  considered 
when  characterizing  a  single  train  of  echoes,  but 
one  or  more  particular  features  of  each  ACF  may 
be  extracted  in  order  to  summarise  quantitatively 
any  given  set  of  echo  samples.    For  example,  the 
positions  (and  to  a  lesser  extent  the  shapes)  of 
peaks  within  the  ACF  indicate  distances  over  which 
there  exists  structural  coherence;  prominent  and 
regular  peaks  indicate  a  regular  arrangement  of 
scattering  sites.    The  mean  number  of  crossings, 
moments  of  the  function,  or  curve  fitting  tech- 
niques using  small  numbers  of  parameters,  may  each 
be  used  to  describe,  concisely,  essential  details 
of  an  estimated  ACF,  but  the  method  of  feature  ex- 
traction chosen  will  depend  on  the  particular  ap- 
pl ication. 

In  order  to  overcome  some  of  the  objections 
that  have  been  indicated  above  against  the  use  of 
single  realisations  of  echo  characteristics  in 
one  dimension,  the  analysis  has  been  restricted 
to  situations  where  the  tissue  investigated  has 
been  found  to  be  isotropic,  or  where  only  the 
changes  with  time  of  echo  characteristics  are 
deemed  to  be  clinically  useful.    In  this  way,  by 
following  the  time  course  of  echo  features,  each 
tissue  acts  as  its  own  internal  standard  and  less 
reliance  need  be  placed  on  the  values  of  individu- 
al, single  measurements.    Time-plots  of  echo 
features  such  as  those  described  above  depict 
changes  in  scattering  structure,  whilst  polar 
phase  diagrams  are  useful  when  the  change  is 
periodic.    Cross  correlation  of  the  ACFs  realized 
at  different  times  are  calculated  to  yield  infor- 
mat'lon  about  changes,  relative  to  a  chosen  ref- 
erence time,  and  examples  are  given  below. 

In  analysing  echoes  in  this  fashion,  the  ACF 
features  to  be  emphasized  have  often  to  be  chosen 
arbitrarily,  and  caution  has  to  be  exercised  when 
attaching  significance  to  any  particular  one. 


277 


There  are  some  reliable  guidelines  that  may  be 
derived  bv  standard  techniques  to  suggest  which 
ACF  values  may  be  regarded  as  spurious,  but  a 
final  decision  as  to  the  significance  (or  other- 
wise) of  any  derived  result  can  be  made  only  on 
the  basis  of  clinical,  in  vivo  investigations. 
However,  as  an  interim  indicator  of  our  aware- 
ness of  artifactual  correlations,  the  results 
given  below  are  marked  with  the  95  percent  con- 
fidence limits  (a)  for  an  autocorrelation  func- 
tion generated  with  N  data  points  from  a  random 
process  [11], 


5.  Results 

There  are  several  clinical  problems  which  may 
be  explored  by  using  temporal  characteristics  of 
the  ACFs  of  echoes  to  indicate  changes  within  tis- 
sues and  three  classes  may  be  distinguished. 

First,  there  are  cyclic  events  in  which  echo 
characteristics  alter  in  synchronism  with  some 
physiologically  significant,  repetitive  process, 
such  as  respiration,  or  with  the  cardiac  cycle, 
whose  phase  is  easily  monitored  on  the  electro- 
cardiograph (ECG).    It  is,  for  example,  reasonable 
to  suppose  that  myocardial  activity  will  be  asso- 
ciated with  a  redistribution  of  scattering  centres 
in  the  tissue,  since  it  necessarily  involves 
changes  in  both  its  elastic  state  and  density. 
The  time  course  of  scattered  echoes  from  within 
the  left  ventricular  posterior  wall  is  easily  fol- 
lowed in  the  conventional  mitral  valve  echogram 
scanning  direction.    Figure  1  depicts  the  ACFs 
measured  from  a  normal  heart  in  end-systole  and 
end-diastole,  showing  typical  changes  from  one 
phase  to  the  other.    The  interpretation  of  this  in 


MM 


Fig.  1.    The  ACFs  of  echoes  from  the  left  ventricu- 
lar posterior  wall  of  a  normal  heart,  mea- 
sured in  end-systole  (S)  and  end-diastole 
(D). 

terms  of  alterations  in  muscle  fibre  dimensions  or 
elastic  state  remains  to  be  confirmed,  whilst  the 
degree  of  change  in  disease  remains  to  be  estab- 
lished, but  the  possibility  clearly  exists  of  mea- 
suring in  vivo  characteristics  which  may  be  relat- 
ed to  myocardial  contractility. 

The  flow  of  blood  in  the  systemic  circulation 


is  also  directly  coupled  with  cardiac  events. 
Following  the  notion  that  the  perfusion  of  cer- 
tain organs  with  blood  proceeds  in  synchronism 
with  the  heart's  action,  the  possibility  of  identi- 
fying poorly  perfused  transplanted  kidneys  is 
under  investigation.    Figure  2  indicates  how 
echoes  from  the  anterior  region  of  the  upper  pole 
of  a  viable  kidney,  investigated  some  months  after 
successful  transplantation,  change  with  time.  The 
obvious  changes  between  phases,  and  the  reversion 
in  the  subsequent  cycle,  is  suggestive  of  a  cyclic, 
cardiac-synchronised  structural  reorganisation 
within  the  kidney.    This  approach  to  the  quantita- 
tion of  echo  characteristics  may  also  be  applica- 
ble to  other  organs,  such  as  the  liver  and  pla- 
centa, and  possibly  to  tumours. 


Fig.  2.    Characterisation  of  echoes  from  cortex  of 
transplanted  kidney  at  different  times. 
Three  autocorrelations  were  measured,  at 
(cardiac)  end-diastole,  end-systole,  and 
the  subsequent  end-diastole.    Curve  A  is 
the  cross-correlation  of  the  first  diastol- 
ic ACF  with  itself,  curve  B  is  the  cross- 
correlation  of  the  first  diastolic  and 
systolic  ACFs,  whilst  curve  C  is  the  cross- 
correlation  of  the  two  diastolic  functions, 
and  shows  a  reversion  back  to  the  original 
state  in  synchronism  with  the  heart. 

A  different  cyclic  process  with  less  regular 
period  is  the  muscular  activity  associated  with 
gastric  emptying.    The  contraction  of  the  stomach 
wall  in  normal  digestion  (which  may  be  monitored 
over  cycles  with  periods  of  approximately  20 
seconds  with  conventional  M-  and  B-scan  tech- 
niques) [12]  produces  changes  in  echo  charac- 
teristics not  dissimilar  to  those  associated  with 
myocardial  contraction,  and  this  lends  support  to 
our  interpretation  of  the  origin  of  heart  wall 
echo  changes.    Thus,  a  similar  model,  viz. ,  con- 
traction of  muscle  elements  causing  changes  to 
their  acoustic  scattering  properties,  may  be  em- 
ployed to  interpret  the  results  of  both  cardiac 
and  gastric  measurements.    ACFs  of  echoes  from 
the  anterior  stomach  wall  are  shown  in  figure  3, 
and  cross  correlations  of  the  ACFs  from  different 
phases  in  figure  4:    the  position  of  the  first 
principal  minimum  is  plotted  as  a  rotating  phasor 
in  figure  5. 


278 


Fig.  3.    The  ACFs  of  echoes  from  the  anterior 

stomach  wall,  contracted  (C)  and  relaxed 
(R). 


Fig.  5.  Phasor  diagram  showing  locus  of  the  posi- 
tion of  the  first  minimum  of  the  function 
produced  by  cross-correlating  the  ACFs  of 
echoes  from  stomach  wall  measured  at  dif- 
ferent times  (as  in  fig.  5).  Total  cycle 
time  20  seconds.  The  radius  of  the 
circle  corresponds  to  a  correlation  lag 
of  4.8  mm. 


.5- 


MM 


Fig.  4.    Characteristics  of  echoes  from  stomach 

wall  at  different  times.    The  ACF  at  time 
0  is  cross-correlated  with  itself  (A),  with 
the  ACF  at  a  time     4  seconds  later  when 
the  stomach  is  contracted  (B),  and  with 
the  ACF  at  a  time     8  seconds  after  con- 
traction, when  the  muscle  has  again  re- 
laxed (C).    These  events  were  chosen  by 
reference  to  the  stomach  M-scan.  Note 
the  general  shape  of  the  cross-correlation 
function  does  not  change  but  the  first 
minimum  moves  with  contraction. 

A  second  class  of  time  change  being  investigat- 
ed is  the  alteration  in  scattering  characteristics 
of  tissues  with  age  or  maturity,  and  of  particular 
interest  are  changes  in  the  placenta  throughout 
pregnancy.    Ageing  placentas  present  different  grey 
scale  B-scan  appearances:    cross-correlation  of  ACFs 
obtained  at  different  stages  of  normal  pregnancy 
are  shown  in    figure  6.    This  technique  may  also 
reveal  differences  between  normal  and  insufficient 
placentas  of  the  same  maturity. 

A  third  class  of  application  is  the  monitoring 
of  structural  changes  in  tumours  during  and  after 
treatment.    Such  changes  have  already  been  observ- 
ed in  B-scan  images  of  irradiated  malignant  tumours 
and  in  vivo  ultrasonic  tissue  characterization 
using  the  techniques  outlined  above  may  provide  a 
clinically  useful  method  of  assessing  tumour  kine- 
tics [13]. 


.5- 


1                                      '  \ 

\      f\  \ 

\ 
i\ 

\\        .7         \             /  , 

V  2/         \.  /6 

V  //      \  / 

V    '  MM 

\*    /  /             V  R  7 

■  1 
1     '/  \ 

C/ 

_i — / ,  \  -usy  V'v 

\      /  2    \    /  4  •'.^      /  \  ' 

1/                  \    1                          ^<k»''  \ 

Fig.  6.    Characteristics  of  echoes  from  placenta 
at  different  stages  of  pregnancy.  ACFs 
were  obtained  at  15,  24,  38  and  41  weeks, 
and  each  of  these  ACFs  was  cross-correlat- 
ed with  the  15  week  ACF  to  produce  curves 
A,  B,  C  and  D  respectively.    There  appears 
to  be  little  change  between  15  and  24 
weeks,  or  between  38  and  41  weeks,  where- 
as there  is  a  clear  change  between  24  and 
38  weeks. 


279 


6.  Conclusions 

The  use  of  single  autocorrelation  functions  to 
characterize  complex,  three-dimensional  scattering 
media  may  be  severely  limited  because  different 
tissues  may  realise  similar  results.    Many  of  the 
objections  raised  against  the  use  of  one-dimen- 
sional techniques,  however,  are  overcome  if  the 
time  course  of  events  is  followed  and  only  changes 
in  echo  characteristics  are  regarded  as  signifi- 
cant.   Whilst  the  ACF  sunsnarises  the  information 
contained  in  a  single  train  of  echoes,  the  cross 
correlation  of  sequentially  produced  ACFs  sum- 
marises dynamic  features  of  scattering.    New  diag- 
nostic information  may  possibly  be  derived  from 
a  study  of  cyclic  changes  in  gastric  and  cardiac 
muscle,  and  in  other  organs  blood  perfusion  may 
be  assessed.    Changes  in  the  nature  of  tissues 
with  age  or  treatment  may  have  important  signifi- 
cance which,  as  yet,  we  are  unable  to  evaluate. 

A  limited  number  of  patients  and  volunteers 
have  so  far  been  investigated  with  this  "dynamic" 
technique,  and  it  is  impossible  to  unequivocally 
claim  success  in  any  particular  application;  but 
there  is  equally  no  cause  to  reject  the  original 
concept  that  echo  ACFs  reveal  changes  in  the 
physiological  state  of  tissues.    The  influence  of 
factors  such  as  tissue  movement  or  anisotropy  and 
the  relative  merits  of  different  transducers  and 
methods  of  signal  processing  will  have  to  be  care- 
fully evaluated,  and  the  optimal  choice  of  echo 
ACF  features  to  be  extracted  remains  uncertain. 
Despite  the  problems,  there  is  every  reason  to 
hope  that  dynamic  tissue  characterization  is  an 
appropriate  path  to  obtaining  information  of 
genuine  clinical  utility. 

Acknowledgements 

We  gratefully  acknowledge  Mr.  George  Hooker's 
help  with  the  analysis  of  data,  and  thank  Dr. 
E.  W.  Emery  for  his  useful  discussions  and  com^ 
ments.    S.  L.  is  in  receipt  of  a  Wellcome  Fellow- 
ship. 

References 

[1]    Lizzi,  F.,  Katz,  L.,  St.  Louis,  L.,  and 
Coleman,  D.  J.,  Applications  of  Spectral 
Analysis  in  Medical  Ultrasonography, 
Ultrasonics,  77-80,  March  1976. 

[2]    Trier,  H.  G.,  Decker,  D.,  Reuter,  R. , 
Epple,  E.,  Lepper,  R.  D.,  and  Nagel ,  M. , 
Frequency-modulated  Portions  of  the  Time- 
amplitude  Ultrasonogram  of  Models,  in 
Proceedings  of  the  Second  European  Congress 
on  Ultrasonics  in  Medicine,  pp.  121-128 
(Excerpta  Medica,  Amsterdam-Oxford,  1975). 

[3]  Gore,  J.  C.  and  Leeman,  S.,  Echo  structure 
in  medical  ultrasonic  pulse-echo  scanning, 
Phys.  Med.  Biol.  22^  (3),  431-443  (1977). 

[4]    Morse,  P.  M.  and  Ingard,  K.  N.,  Theoretical 
Acoustics  (McGraw-Hill,  New  York,  1968). 

[5]    Leeman,  S. ,  Simple  Physical  Ideas  on  Ultra- 
sound Pulse  Scattering  from  Tissue,  R.P.M.S. 
Medical  Physics  Reports,  US76/3  (1976). 
(Available  by  request  only) 


[6]    Waag,  R.  C.  and  Lerner,  R.  M.,  in  Ultra- 
sonics Symposium  Proceedings,  I . E . E . E . 
Cat.  73,  CHO  807-850. 

[7]    Gore,  J.  C.  and  Leeman,  S.,  Ultrasonic  back- 
scattering  from  human  tissue:    A  realistic 
model,  Phys.  Med.  Biol.  22  (2),  317-326 
(1977). 

[8]    Gore,  J.  C.  and  Leeman,  S.,  New  Criteria 
for  the  Assessment  of  the  Resolution  of 
Ultrasonic  Scanners,  in  Ultrasonics  in 
Medicine,  E.  Kazner,  M.  de  Vlieger,  H.  R. 
Muller,  and  V.  R.  McCready,  eds.,  pp.  197- 
203  (Excerpta  Medica,  Amsterdam,  1975). 

[9]    Fields,  S.  and  Dunn,  F. ,  J.  Acoust.  Soc. 
Am.  54,  809-813  (1973). 

[10]    Jenkins,  G.  M.  and  Watts,  D.  G.,  Spectral 
Analysis  and  its  Applications  (Holden-Day, 
San  Francisco,  1968). 

[11]    Chatfield,  C. ,  The  Analysis  of  Time  Series: 
Theory  and  Practice,  Chap.  4  (Chapman  and 
Hall,  London,  19/b). 

[12]    Bateman,  D.  N.,  Leeman,  S.,  Metreweli,  C, 
and  Willson,  K. ,  A  noninvasive  technique 
for  gastric  motility  measurement,  Brit.  J. 
Radiol.  50,  526-527  (1977). 

[13]    Leeman,  S.,  Badcock,  P.  C,  Gore,  J.  C, 
Plessner,  N.  J.,  and  Willson,  K.,  Ultra- 
sonic Backscattering  Assessment  of  Tumour 
Response  to  Treatment,  presented  to  Tumour 
Ultrasound  77,  International  Conference, 
London,  1977.    (Copy  available  by  request.) 


280 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


COMPUTER  SPECTRAL  ANALYSIS  OF  ULTRASONIC  A  MODE  ECHOES 


D.  E.  Robinson 

Ultrasonics  Institute 
Sydney,  Australia 


Tissue  attenuation  as  a  function  of  frequency  measured  by  the  effect  on  the  power 
density  spectrum  of  the  echo  off  glass  shadowed  by  the  tissue  sample  is  shown  to  give 
results  comparable  to  those  published  previously.    A  method  of  measuring  attenuation 
within  tissue  by  comparison  of  the  scattered  echoes  from  shallow  and  deep  scatterers 
is  investigated  and  shown  to  have  limitations.    An  assessment  of  properties  of  the 
scatterers  from  investigation  of  the  spectral  properties  of  the  echoes  from  a  scatter- 
ing region  is  suggested. 


Key  words:    Acoustic;  computer  processing;  digital  acquisition;  digital  signal 
processing;  pulse-echo  techniques;  spectrum  analysis;  ultrasonics. 


1.  Introduction 

At  present  the  bulk  of  tissue  characterisation 
by  ultrasound  in  clinical  practice  is  done  by 
qualitative  examination  of  the  grey  scale  B  Mode 
echogram.    The  current  work  is  directed  to  deriv- 
ing quantitative  characterisation  data  by  spec- 
tral analysis  of  180°  backscattered  echoes,  or 
ultrasonic  A  Mode,  following  the  approach  of 
Lizzi  [l]i. 

2.    Measuring  Equipment 

The  transducer  beam  pattern  requires  to  be  nar- 
row to  give  good  spatial  resolution  and  with  a 
well  behaved  spectrum  to  avoid  complications  in 
spectral  analysis  techniques  [2].    This  is  achiev- 
ed in  the  focal  zone  of  a  focussed  transducer  of 
aperture  65  mm  and  curvature  35  cm  and  usable  fre- 
quency range  from  1.5  to  4  MHz.    The  20  dB  echo 
beamwidth  at  the  focus  is  6.7  mm. 

The  receiver  amplifier  gain  is  variable  in  1  dB 
steps  allowing  an  input  dynamic  range  of  100  dB 
without  overload.    Receiver  gain  variation  is  used 
instead  of  transmitter  power  control  as  it  is  found 
that  due  to  non-linearities  in  the  transmitter,  the 
transducer  or  the  medium,  significant  changes  in 
echo  wave  shape  occur  with  changes  in  transmitter 
power.    The  signals  are  digitised  at  10  MHz  sam- 
pling rate  using  an  8  bit  Biomation  8100  waveform 
recorder  interfaced  to  an  Interdata  Model  85  com- 
puter.   The  computer  and  an  interactive  signal 
processing  program  have  been  described  elsewhere 
[3]. 

The  spectrum  used  here  is  the  square  magnitude 
of  the  Discrete  Fourier  Transform  (DFT)  of  the 
echo  signal  or  Power  Density  Spectrum.    All  spectra 
shown  in  this  paper  have  been  interpolated  as  de- 
scribed by  Rabiner  et  al.  [4]  by  extending  the  set 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


of  128  or  256  recorded  time  samples  with  zero- 
valued  samples  up  to  a  total  signal  length  of  1024 
before  computing  the  DFT.    Comparisons  between 
spectra  are  made  by  calculating  the  logarithm  of 
the  ratio  of  the  spectra  (log  ratio)  and  observing 
the  ordinate  at  2.5  MHz  in  dB  and  the  slope  in 
dB/MHz.    The  log  ratio  for  point  and  plane  target 
echoes  remain  within  3  dB  and  the  slope  within 
2  dB/MHz  for  distances  from  the  transducer  focus 
of  ±  30  mm. 

A  B-Mode  display  of  the  examined  tissue  is  pro- 
vided for  A  mode  beam  position  selection  and  guid- 
ance.   An  example  of  input  data  and  corresponding 
B  mode  display  is  shown  in  figure  1. 

■  3.    Experimental  Techniques 

First  the  echo  from  a  thick  glass  block  at  the 
focus  with  no  intervening  tissue  is  recorded,  and 
regarded  as  the  impulse  response  of  the  system. 
The  tissue  specimen  is  then  placed  in  a  poly- 
ethylene bag  in  front  of  the  glass  block  and  echoes 
from  a  series  of  30  adjacent  beam  positions  at  3  mm 
intervals  were  recorded  from  three  locations;  from 
the  glass  block  behind  the  tissue  and  from  shallow 
and  deep  within  the  examined  tissue.    In  each  case 
the  transducer  was  moved  so  that  the  echoes  were 
recorded  from  the  focal  area  of  the  transducer 
beam.    In  the  case  of  the  scattered  echoes,  a 
region  was  selected  which  did  not  have  large  dis- 
crete echoes  on  the  B  Mode  scan. 

4.    Signal  Processing 

When  measuring  the  spectrum  of  a  statistical  sig- 
nal, some  form  of  averaging  is  required  to  obtain  a 
smooth  and  stable  spectral  estimate  [5].    This  can 
be  done  in  the  spatial  domain  by  translating  the  beam 
axis  as  described  by  Lizzi  [1]  and  adding  the  spectra 
corresponding  to  a  number  of  adjacent  lines  of  sight 
to  obtain  an  average  spectrum.    This  has  the  effect 
of  reducing  the  amplitude  of  ripples  in  the  spectrum. 


281 


Fig.  1.    B-Mode  echogram  showing  tissue  sample 
in  front  of  glass  block.    The  two 
lines  indicate  the  area  used  for  A 
Mode  acquisition.    The  complete  A  mode 
signal  and  the  part  of  the  signal  near 
the  focal  region  are  also  shown. 


It  should  be  noted  that  the  "fine  structure"  in  the 
spectrum  of  individual  echo  signals  may  be  a  func- 
tion, not  of  the  statistical  nature  of  the  sample, 
nor  of  the  local  tissue  structure,  but  simply  of  the 
length  of  the  time  window.    For  instance,  given  a 
signal  x(t)  with  transform  X(f)  and  spectrum  X(F)2 
the  sum  of  two  identical  shifted  signals  x(t)  + 
x(t  -  t)  has  a  spectrum  2X(f)2  (1  +  cos  wt).  The 
factor  (1  +  cos  wt)  gives  rise  to  "wiggles"  in  the 
measured  spectrum  which  are  not  directly  related  to 
local  tissue  "structure"  but  are  merely  phase  ef- 
fects between  widely  spaced  echoes.    Figure  2a  and 
2b  show  the  spectra  for  a  single  echo  and  two  identi- 
cal echoes  separated  by  5  mm.    This  effect  is  re- 
duced by  windowing  in  the  autocorrelation  domain,  or 
lag  windowing.    The  autocorrelation  function  is  the 
Inverse  Fourier  Transform  of  the  Power  Density  Spec- 
trum.   The  Autocorrelation  functions  of  the  single 
echo  and  two  identical  echoes  spaced  at  5  mm  (66 
sample  intervals)  are  shown  in  figure  2c  and  2d. 
Also  shown  in  figure  2d  is  a  Hamming  window  in  the 
Autocorrelation  (lag)  domain  suitable  for  removing 
the  effect  of  widely  spaced  echoes.  Multiplication 
by  the  window  function  has  the  effect  of  removing 
contributions  to  the  Autocorrelation  function  of 
lags  greater  than  32  sample  intervals  (2.5  mm)  and 
leaving  only  the  contributions  from  closely  spaced 
echoes.    The  smoothed  Power  Density  Spectrum  ob- 
tained by  re-transforming  the  weighted  Autocorrela- 
tion function  into  the  frequency  domain  is  shown  in 
figure  2e.    The  effect  of  this  procedure  on  tissue 
echoes  is  shown  in  figure  3  where  the  original  time 
sample  of  scattered  echo  data  is  windowed  with  a 
Hamming  window  of  length  128  (corresponding  to  10  mm 


of  tissue).    The  lag  windows  are  of  length  64,  32 
and  16  samples  increments  corresponding  to  correla- 
tion lengths  of  5,  2.5  and  1.25  mm  and  thus  limit- 
ing the  effect  on  the  spectrum  to  structures  of 
these  sizes. 

In  the  analysis  of  the  echoes  off  glass  no  time 
windowing  or  spectral  smoothing  is  necessary  since 
we  are  dealing  with  only  deterministic  single  echo 
waveforms  and  not  statistical  signals.    The  power 
density  spectra  of  the  echo  off  glass  with  and  with- 
out shadowing  by  the  tissue  were  calculated,  and  the 
log  ratio  formed.    A  typical  set  of  results  is  shown 
in  figure  4.    The  log  ratio  is  a  measure  of  the  at- 
tenuation due  to  the  tissue  as  a  function  of  fre- 
quency.   The  values  are  only  significant  for  fre- 
quencies contained  in  the  original  power  density 
spectrum  of  the  unshadowed  glass  echo.    A  line  of 
best  fit  was  computed  on  a  least  total  weighted 
squared  error  basis  with  the  weighting  function  be- 
ing the  original  power  density  spectrum  values  for 
the  echo  off  glass  at  the  focus. 

The  parameters  of  the  line  y  =  ax  +  b  where  y  is 
the  best  fit  to  the  log  ratio  value  and  x  the  fre- 
quency value  are  given  by: 


a  = 


?  W.  ?  x.  y.  W.  -  ?  x.  W.  ?  y.  W. 
1      11      1-^11       1  11111 


and  b  = 


?    W.    ^    x?  W. 
?       W.      ^   x.  W. 


?  w. 

1  1 


?  w. 

1  1 


282 


Fig.  2.    a)  Power  density  spectrum  of  a  single 
echo  off  glass;  b)  spectrum  of  two 
echoes  separated  by  .5  mm;  c)  auto- 
correlation function  corresponding  to 
spectrum  in  a;  d)  autocorrelation 
function  of  spectrum  shown  in  b  and 
64  point  Hamming  lag  window; 
e)  smoothed  spectrum  obtained  from 
the  lag  windowed  autocorrelation 
function  shown  in  d. 


where  the  x.,  y.  and  W.  are  the  values  of  the  fre- 
quency, logVatIo  and  vveighting  function  for  the 
successive  frequency  samples. 

For  echoes  from  scatterers  within  the  tissue 
the  echo  signal  was  windowed  using  a  time  window 
of  length  128  samples  or  10  mm  and  the  spectrum 
was  smoothed  by  a  lag  window  of  64  lags  correspond- 
ing to  5  mm.    The  log  ratio  between  deep  and  shal- 
low echo  spectra  was  formed  and  the  weighted  best 
fit  line  calculated  as  before.    A  typical  result  is 
shown  in  figure  5. 


5.  Results 

Being  initial  experiments  readily  available  ani- 
mal material  was  used.    The  material  was  skeletal 
beef  muscle  along  and  across  the  fibre  direction 
and  calf  liver.    The  results  are  shown  in  table  1. 

The  log  ratio  curve,  indicating  tissue  attenua- 
tion was  approximately  linear  within  the  pass  band 
for  all  tissue  samples  using  the  shadowed  glass  ap- 
proach.   For  this  reason  more  confidence  is  placed 
in  the  determination  of  attenuation  slope  by  this 


283 


Fig.  3.    Effect  of  lag  windowing  on  echo  spectrum.    A  single  line  of  data  was  windowed 
with  a  1  cm  Hamming  window  and  the  spectrum  a)  smoothed  with  a  Hamming  lag 
window  of  b)  5  mm,  c)  2.5  mm  and  d)  1.25  mm. 


Table  1.    Tissue  attenuation  determined  by  the  shadowed  glass 
method  and  by  the  scattered  echo  comparison  method. 


Tissue         Quoted  atten.  Shadowed  glass  Scattered 

(Wells  [6])  Slope      Atten.  at  2.5  MHz     Slope     Atten.  at  2.5  MHz 

dB/cm  MHz         dB/cm  MHz  dB/cm  dB/cm  MHz  dB/cm 


Muscle  along 


f i  bres 

1.3 

1.5 

4  .9 

1.2 

5.8 

Muscle  across 

f i  bres 

3.3 

(1) 

.65 

1.5 

.28 

2.8 

(2) 

.78 

1.8 

.15 

1.7 

Li  ver 

.94 

(1) 

.82 

2.2 

.72 

3.0 

(2) 

.73 

3.1 

.53 

4.2 

284 


Fig.  4.    Results  of  shadowed  glass  measurement  on  muscle  across  fibre  direction  showing 
a)  unshadowed  and  b)  shadowed  spectra,  c)  ratio  of  spectra  and  d)  log  ratio 
with  weighted  best  fit  line  superimposed.    Note  that  the  ripple  amplitude  is 
small  in  the  linear  region. 


method.    The  values  attenuation  versus  frequency 
in  dB/cm  MHz  were  in  reasonable  agreement  with  pre- 
viously published  values  for  the  muscle  along  the 
fibres  and  for  liver.    The  values  for  muscle  across 
the  fibres  were  similar  to  the  other  values  but  not 
in  agreement  with  Wells  [6].    The  log  ratio  curves 
for  the  scattered  echo  measurements  all  showed 
significant  ripples,  indicating  that  some  phasing 
effects  still  remain.    The  values  for  liver  and 
muscle  along  the  fibres  agreed  reasonably  well  with 
the  shadowed  glass  method,  but  the  muscle  along 
fibres  gave  widely  different  results. 

6.  Discussion 

It  is  obvious  that  further  work  is  needed  before 
the  scattered  echo  comparison  method  will  be  useful 
for  attenuation  determination.    The  shadowed  glass 
method  appears  to  give  repeatable  results  but  is  not 
suitable  for  clinical  examinations. 

The  current  approach  in  scattered  echo  analysis 
is  to  average  in  the  power  spectral  domain  with  the 
aim  of  reducing  the  ripples  in  the  spectrum  to  ob- 
tain a  smooth  and  stable  spectral  estimate.    It  ap- 


pears that  an  alternative  approach  is  to  average 
in  another  domain  to  obtain  a  stable  estimate  of 
the  underlying  spectral  ripple,  which  is  a  measure 
of  the  "structure"  of  the  echo  signal.    Two  obvious 
candidates  are  the  Autocorrelation  and  the  cepstral 
domains.    This  structure  can  be  postulated  from  the 
same  model  of  distributed  discrete  point  scatterers 
that  is  used  to  justify  the  single  frequency  scat- 
tering versus  angle  approach  to  tissue  characteri- 
sation . 


References 

[1]    Lizzi,  F.,  Katz,  L.  ,  St.  Louis,  L.,  and  Cole- 
man, D.  J.,  Applications  of  spectral  analysis 
in  medical  ultrasonography.  Ultrasonics  14,  77- 
80  (1976). 

[2]    Robinson,  D.  E.  and  Williams,  B.  G.,  Computer 
Analysis  of  Ultrasonic  Pulse  Echo  Signals,  in 
Ultrasound  in  Medicine  and  Biology,  D.  White 
and  R.  E.  Brown,  eds.,  p.  1443  (Plenum  Press, 
New  York  1977). 


285 


6db 

j  M  1 

* 

m 

Fig.  5.    Results  of  scattered  measurement  showing  a)  shallow  and  b)  deep  echo  spectra, 
c)  the  ratio  and  d)  log  ratio  with  best  fit  line  superimposed.    Note  the 
large  ripple  content  of  the  log  ratio. 


1  n 


[3]    Robinson,  D.  E.,  Williams,  B.  G.,  and  Horn, 
P.  R.,  Digital  acquisition  and  interactive 
processing  of  ultrasonic  echoes.  Ultrasound 
Med.  S  Biol .  2,  199-212  (1976). 

[4]    Rabiner,  L.  R.  ,  Gold,  B.,  and  McGonegal,  C.  A. 
An  Approach  to  the  Approximation  Problem  for 
Nonrecursive  Digital  Filters,  in  IEEE  Trans. 
Audio  Electroacoustics ,  AV-IS,  83-106  (  1970). 


[5] 


[6] 


Jenkins,  G.  M.  and  Watts,  D.  G.,  Spectral 
Analysis  and  its  Application,  p.  209  (Holder- 
Day  1968). 


Wells,  P.  N.  T. , 
sonic  Diagnosis. 


Physical  Principles  of  Ultra- 
Academy  Press  1969. 


286 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


CEPSTRAL  SIGNAL  PROCESSING  FOR  TISSUE  SIGNATURE  ANALYSIS 


J.  Eraser  and  G.  S.  Kino 

Stanford  University 
Stanford,  California    94305,  U.S.A. 

J.  Birnholz 

Harvard  Medical  School 
Boston,  Massachusetts    02115,  U.S.A. 


The  reflected  signal  received  by  an  ultrasonic  transducer  is  modeled  as  a  convolu- 
tion of  a  transducer  response  with  a  reflection  function  for  the  target  region. 
Cepstral  analysis  translates  that  signal  into  a  domain  where  those  components  inter- 
act additively  rather  than  through  convolution  and  where  separation  can  be  accomplish- 
ed with  simple  bandpass  filtering  operations.    The  cepstral  transform  of  the  pulse 
echo  signal  also  provides  direct  access  to  any  periodic  behavior  Ci'  reflectors  aris- 
ing from  their  spacing.    As  an  example  of  these  capabilities,  the  technique  is  used 
experimentally  to  describe  the  characteristic  spacing  of  macrostructural  reflecting 
elements  in  the  pig  liver  in  vitro  and  to  determine  the  frequency  dependent  attenua- 
tion behavior  of  normal  human  liver  in  vivo. 


Keywords:    Attenuation;  B-scan;  cepstrum;  computer;  convolution;  de-convolution; 

liver;  power  spectrum;  signal  processing;  tissue  characterization; 
tissue  parameters;  ultrasound. 


1.  Introduction 

We  describe  in  this  paper  preliminary  experi- 
ments with  new  types  of  non-linear  signal  proces- 
sing used  for  eliminating  certain  types  of  signal 
distortion  in  conventional  acoustic  imaging  sys- 
tems which  preclude  unambiguous  identification  of 
specific  tissue  features. 

An  acoustic  propagation  path  through  a  tissue 
region  can  be  treated  as  a  one-dimensional  array 
of  reflectors.    While  each  individual  reflector 
may  be  expected  to  replicate  the  time  charac- 
teristics of  a  self  convolved  transducer  impulse 
response  (assuming  an  impulsive  excitation),  that 
pulse  form  is  altered  or  distorted  during  propaga- 
tion by  the  frequency  dependent  attenuation 
processes  of  the  medium  intervening  between  re- 
flector and  transducer.    Previous  studies  have  in- 
dicated that  tissue  specific  information  is  de- 
rived from  both  attenuation  properties  of  the 
region,  as  well  as  the  distribution  and  backscat- 
ter  features  of  the  individual  reflectors  [1]^. 
Therefore,  an  effective  signal  processing  scheme 
must  provide  a  means  of  eliminating  the  trans- 
ducer response  and  separating  frequency  dependent 
attenuation  effects  from  target  signal  components. 
Our  approach  has  been  to  take  advantage  of  certain 
properties  of  the  cepstrum,  or  Fourier  transform, 
of  the  signal  log  power  spectrum  [2].    The  ration- 
ale for  this  choice  derives  from  previous  uses  in 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


comparable  signal  processing  tasks:    (1)  homo- 
morphic  deconvol ution  of  speech  and  music  wave- 
forms for  improved  recording  fidelity  [3];  (2) 
dereverberation  enhancement  of  seismic  data  [4]; 
and  (3)  recognition  and  characterization  of  flaw 
configuration  in  nondestructive  testing  [5]. 
This  method  has  not  been  applied  to  medical  diag- 
nostic ultrasound  previously. 

2.    Analysis  of  the  Return  Echoes 

We  consider  first  what  occurs  when  a  signal  is 
incident  on  body  tissue.    We  will  mainly  be  in- 
terested in  the  situation  in  which  we  observe  a 
reflected  signal  with  the  same  transducer  used  to 
iTluminate  the  tissue.    Techniques  for  looking  at 
off  axis  reflections  can,  in  principle,  provide  a 
great  deal  more  information.    But,  in  practice, 
they  are  extremely  difficult  to  employ  without  the 
use  of  arrays  of  transducers  to  carry  out  repeat- 
able  measurements  of  this  kind.    Even  with  the  use 
of  arrays  of  transducers,  because  of  the  distor- 
tions along  the  body  path  to  an  organ  of  interest, 
the  problems  are  very  severe  in  terms  of  repeat- 
ability.   In  that  case,  it  is  probably  better  to 
concentrate  on  the  development  of  imaging  systems 
first,  then  process  the  images  in  some  detail. 
Thus,  the  method  described  here  is  entirely  devot- 
ed to  the  use  of  a  single  transducer  operating  in 
a  reflection  mode. 

The  signal  received  by  a  reflection  mode  system 
can  be  considered  to  arise  from  a  one-dimensional 
reflection  function  R  of  distance  z  along  the  beam 
path  R(z).    In  the  far  field  of  a  plane  transducer. 


287 


or  the  focal  zone  of  a  focused  transducer,  the 
phase  fronts  are  roughly  planar  and  normal  to  the 
beam.    Under  these  conditions,  it  can  be  shown  by 
perturbation  theory  that  R(z)  is  the  weighted  sum 
across  the  beam  of  the  component  along  the  beam 
of  the  gradient  of  the  acoustic  impedance  of  the 
tissue.    That  is,  if  the  beam  pattern  is  P(r),  and 
the  impedance  is  Z(r,z)  which  varies  slightly 
around  an  average  value  of  Zg, 

R(z)  -       y  2TTrdrP{r)z  •  v|z(r,z)f   .  (D 

°  0 

We  wish  to  work  in  the  time  domain,  using  R(t), 
where,  with  V  the  acoustic  velocity  in  tissue 

t  =  2z/V  (2) 

since  the  time  required  is  for  a  round  trip.  Any 
information  about  R(t)  which  can  be  obtained  will 
be  related  to  changes  of  impedance  in  the  tissue, 
and,  hence,  may  be  of  use  in  tissue  characteriza- 
tion. 

A  region  of  tissue  is  interrogated  by  passing 
a  pulse  A(t)of  ultrasound  through  it  and  observ- 
ing the  return  signal  S(t).    S(t)  is  given  by  the 
convolution  of  A(t)  and  R(t): 

S(t)  =  /A(t)  R(t  -  T)dt  (3) 

which  can  be  represented  symbolically  as 

S{t)  =  A(t)  *  R(t).  (4) 

We  do  not  necessarily  seek  to  recover  R(t)  exact- 
ly, but  rather  to  derive  statistical  information 
about  it.    We  are  interested  in  knowing  the  distri- 
bution sizes  of  the  objects  response  for  the  re- 
flected signal,  and  how  regularly  they  are  spaced 
in  the  tissue.    We  may  also  be  interested  in  the 
evolution  of  A(t)  as  the  pulse  passes  through  tis- 
sue with  frequency  dependent  attenuation. 

Since  the  autocorrelation  function  responds  to 
structures  with  strong  peaks,  it  could  be  con- 
sidered a  possible  processing  technique  to  extract 
structure  information  from  S(t).    However,  it  can 
be  shown  that  the  autocorrelation  of  two  convolved 
functions  is  the  convolution  of  their  autocorrela- 
tions.   Using -A"  to  represent  correlation, 

S(t)*S(t)  =  (A(t)*A(t))  *  (R(t)*R(t))  .  (5) 

Since  A(t)  is  an  oscillatory  function,  A(t)*A(t) 
is  also,  and  the  autocorrelation  of  R  is  confused 
by  the  convolution.    The  best  that  can  be  done  in 
this  case  is  to  recover  the  envelope  of  the  auto- 
correlation.   A  convenient  technique  for  this  pur- 
pose is  to  find  the  envelope  of  the  analytic  func- 
tion associated  with  the  autocorrelation  by  taking 
the  Fourier  transform,  zeroing  negative  frequency 
components,  and  taking  the  magnitude  of  the  inverse 
transform  [6].    Since  the  autocorrelation  is  normal- 
ly performed  by  inverse  transforming  the  power 
spectrum,  very  little  additional  effort  is  needed. 
One  merely  zeros  the  negative  frequencies  in  the 
power  spectrum. 

An  example  of  a  typical  ultrasound  echo  is 
shown  in  figure  1.    This  echo  represents  the  re- 
turn from  a  pig  liver,  using  a  3.5  MHz  transducer. 
A  pattern  of  multiple  echoes  is  evident,  but  not 
readily  quantifiable.    The  envelope  of  the  auto- 


TIME  ,  MICROSECONDS 


Fig.  1.    The  recorded  ultrasonic  echo  from  a 
section  of  fresh  pig  liver,  in  vitro, 
using  a  nominally  3.5  MHz  transducer. 
This  represents  1.9  cm  of  tissue. 


0  12.8 
TIME  ,  MICROSECONDS 


Fig.  2.    The  envelope  of  the  autocorrelation 
function  of  the  signal  in  figure  1. 

correlation  of  this  signal  is  shown  in  figure  2. 
Several  peaks  are  discernable,  but  none  is  dominant. 
A  more  sophisticated  technique  is  sought. 

3.    The  Inversion  Process 

In  tissue  signature  analysis,  responses  from 
the  tissue  are  weighted  by  the  form  of  the  trans- 
ducer response.    In  addition,  there  is  distortion 
due  to  the  tissues  on  the  way  to  the  organ  of 
interest.    So  a  process  is  desired  which,  given 
S(t),  can  separate  A(t)  and  R(t).    Under  ideal 
conditions,  if  one  of  two  convolved  functions  is 
known,  the  other  may  be  found.    Deconvolution  is 
performed  by  Fourier  transforming,  inverse  filter- 
ing, and  inverse  Fourier  transforming. 

If  eq.  (4)  is  Fourier  transformed,  the  convolu- 
tion leads  to  a  product,  and  rearranging  gives: 

RM  =  iM.  .  (6) 

Several  factors  prevent  the  use  of  this  formal  in- 
version process  in  the  present  case.    The  presence 
of  finite  noise  in  the  recorded  signal  means  that 
S(oj)  is  not  known  exactly.    In  addition,  as  the 
ultrasonic  transducer  responses  are  band-limited, 
A(u)  becomes  vanishingly  small  in  the  same  frequen- 
cy ranges  where  S(a))  is  dominated  by  noise,  so  the 
noise  has  a  disproportionate  influence  on  the  re- 
covery of  R(a)).    Finally,  since  the  signal  has  prop 
propagated  through  a  tissue  of  unknown  and  frequen- 
cy dependent  attenuation,  A(a))  is  not  known.  A 
different  approach  is,  therefore  needed. 


288 


4.    Homomorphic  Processing 

We  have  used  so-called  homomorphic  processing, 
or  cepstral  analysis,  for  this  purpose.    At  the 
present  time,  we  have  by  no  means  developed  this 
type  of  processing  to  its  limit,  but  we  have  made 
a  start  on  techniques  employing  it  and  our  inten- 
tion is  to  develop  the  techniques  further. 

The  basic  idea  behind  homomorphic  processing 
is  to  turn  a  nonlinear  process  into  a  linear 
process. 

Consider  the  convolution  to  the  signal  from 
the  transducer  A(t)  with  the  tissue  response 
R(t)  in  the  time  domain 


S(t)  =  A(t)  *  R(t) 


(7) 


In  the  frequency  domain,  the  two  components  de- 
rived from  these  signals  are 


A(a))   •  R(a)) 


(8) 


A  product  form  is  also  obtained  for  the  power 
spectrum 

tS(a,)|2  =    |A(a.)|2   .    |R{^)|2  (9) 

By  taking  the  logarithm  of  |S(q))|2,  the  multipli- 


cation is  converted  to  a  sum.    This  enables 
normal  linear  signal  processing  techniques  to  be 
used. 


Let 


then 


S(a>)  =  log|S(a))|2 

A{o))  =  log|A(a>)|2 

R(a))  =  log|R(a))|2  , 

S(a))  =  A(a))  +  R(oj)  . 


(10) 

(11) 

(12) 
(13) 


Finally,  if  we  take  the  Fourier  transform  of 
S(a))  defined  as  S(t),  we  can  write 


S(t)  =  A(t)  +  R(t) 


(14) 


S(t)  is  called  the  cepstrum  of  S(t).    The  nota- 
tion is  used  to  distinguish  it  from  S(t). 
Figures  3,  4,  and  5  demonstrate  the  convolution 
of  a  simulated  transducer  impulse  response  with 
an  impulse  pair,  and  the  additive  property  of 
the  log  power  spectra  and  cepstra. 

We  note  that  in  the  log  power  spectrum  domain 
the  system  is  linear  and  subject  to  all  the  rules 
and  techniques  of  linear  filtering.    Such  filter- 
ing may  be  accomplished  by  taking  one  more 
Fourier  transform,  yielding  the  cepstrum  of  the 


LjJ 

Q 
Z) 


CL 
< 


CO 


(a) 


(b) 


(c) 


TIME 

Fig.  3.    (a)  A  simulated  transducer  impulse 
response  of  the  form 
g-^(t/8)2  2TT(t/8). 

(b)  An  impulse  pair  with  spacing  16. 

(c)  The  convolution  of  (a)  and  (b). 


A 

A 

FREQUENCY 

Fig.  4.    The  log  power  spectra  of  figure  2(a), 
(b),  and  (c)  respectively. 


289 


(a) 


(b) 


(c) 


TIME 

Fig.  5.    The  cepstra  of  figure  3(a),  (b),  and 
(c)  respectively. 


ceps'trum  of  the  reflection  pattern  is  not  known 
and  varies  from  tissue  to  tissue. 

Some  examples  of  the  cepstra  of  impulse  trains 
can  yield  insight  into  what  may  happen  in  tissue, 
and  suggest  features  which  will  be  meaningful. 
Two  im.pulses  with  separation  t  will  yield  a  string 
of  impulses  with  separation  t  and  of  strength  de- 
caying fairly  rapidly  toward  zero  away  from  the 
origin,  as  seen  in  figures  3(b)  and  5(b).    If  more 
than  two  impulses  exist  with  the  same  spaces,  the 
cepstrum  is  similar,  but  with  a  slower  decay  rate. 
This  is  depicted  in  figure  6.    Figure  6  also  de- 
monstrates a  problem  which  often  arises  in  the 
digital  implementation  of  the  cepstrum.    When  the 
logarithm  of  the  power  spectrum  is  taken,  the  non- 
linear process  generates  harmonics  in  the  time 
domain  which  can  cause  aliasing  in  the  cepstrum, 
even  though  the  original  spectrum  bandwidth  was 
well  within  the  Nyquist  bandwidth  of  the  sampling 
process.    In  general,  any  repetitive  pattern  in  a 
signal  will  lead  to  a  peak  at  a  time  equal  to  that 
spacing  in  the  spectrum,  as  well  as  lesser  peaks 
at  multiples  of  that  spacing.    The  cepstrum  itself 
may  not  be  easy  to  interpret  for  ultrasound  echoes 
from  tissue  unless  the  tissue  has  a  fairly  simple 
and  distinct  structure.    But  the  decay  rate  of  the 
cepstrum  is  related  to  the  spacings  and  numbers  of 
periodic  reflections  in  the  tissue,  and  hence,  may 
be  useful  for  tissue  characterization. 

If  the  minimum  spacing  between  strong  reflec- 
tions in  the  tissue  is  somewhat  greater  than  the 
wavelength  of  the  interrogating  sound,  all  informa- 
tion about  the  periodicities  of  reflections  will 
be  separated  from  the  information  about  the  charac- 
ter of  an  individual  reflection  in  the  cepstrum. 
In  this  case,  it  should  be  possible  to  filter  the 
cepstrum  and  recover  the  average  of  the  power 
spectral  densities  of  the  individual  reflections 
from  a  region  of  tissue.    This  should  be  a  good 


original  signal.    Filtering  may  then  be  perform- 
ed by  multiplying  the  cepstrum  by  a  weighting 
function  to  select  a  desired  portion,  then  re- 
versing as  many  processing  steps  as  necessary  to 
reach  a  useful  and  recognizable  domain.  Since 
information  was  lost  in  taking  the  power  spec- 
trum complete  inversion  is  not  possible.  This 
restriction  is  removed  by  the  use  of  the  complex 
cepstrum,  in  which  the  real  logarithm  of  the 
power  spectrum  is  replaced  by  the  complex  loga- 
rithm of  the  complex  spectrum.    However,  this 
technique,  which  has  been  demonstrated  in  seis- 
mographic  and  speech  processing  applications,  is 
much  more  difficult  to  use  in  practice.    If,  in- 
stead, we  reconstruct  the  signals  from  the  real 
cepstrum,  we  obtain  the  correlation  function  of 
the  original  signals,  or  of  the  tissue  impedance 
variations. 

5.    Possible  Applications 

Filtering  in  the  cepstrum  domain  is  useful  only 
if  the  cepstra  of  the  two  components  fall  in  dif- 
ferent regions  of  the  domain.    The  spectrum  of  a 
broadband  transducer  impulse  response  can  be  close 
to  a  Gaussian  response  in  frequency.    So  its  loga- 
rithm is  close  to  a  parabolic  response  in  fre- 
quency, and  cepstrum  is  a  relatively  narrow  pulse 
centered  at  the  origin,  which  has  a  time  width 
of  the  same  order  as  the  original  impulse.  This 
effect  can  be  seen  in  figures  3(a)  and  5(a).  The 


(a) 


(b) 


TIME 


Fig.  6. 


(a)  Four  impulses  with  spacing  16. 

(b)  The  cepstrum  of  the  convolution 
of  the  impulse  response  of  3(a), 
with  the  impulse  train  of  6(a). 


290 


approximation  to  the  spectrum  of  the  transmitted 
ultrasound  pulse  as  it  passes  through  the  region 
of  tissue  within  an  unknown  constant.    If  such  a 
calculation  is  carried  out  on  signals  from  two 
succeeding  regions  of  tissue  along  the  line  of 
sight,  the  difference  between  them  will  contain 
information  about  the  frequency  dependence  of  the 
attenuation  of  ultrasound  in  the  tissue  between 
these  two  regions. 

6.    Experimental  Techniques 

A  surplus  B-scan  imaging  system,  loaned  by  the 
Radiology  Department  of  the  Stanford  University 
Medical  Center,  was  obtained,  originally  as  a 
means  of  positioning  the  ultrasound  transducers 
and  generating  images  for  identification  of  the 
source  of  echoes  to  be  processed.    It  was  found 
that  the  pulse  generator  and  receiving  amplifier 
of  the  system,  a  Picker,  were  of  excellent  quali- 
ty.   Also,  a  depth  marker  was  available,  which 
could  be  used  to  select  a  portion  of  a  trace. 
By  tapping  a  small  amount  of  the  received  signal 
with  a  high  impedance  isolation  transformer  at  a 
low  impedance  point  in  the  receiver,  a  high  quali- 
ty radio  frequency  signal  was  obtained  without 
compromising  either  the  operation  or  safety  of  the 
existing  B-scan  system.    The  tapped  R-F  signal  was 
further  amplified  and  monitored  by  an  oscilloscope 
and  a  Biomation  8100  transient  recorder,  which 
was  set  to  record  when  armed  by  a  pushbutton  ac- 
cessible to  the  operator  of  the  B-scan  system. 
A  PDPll/10  minicomputer  was  used  to  store  the 
digitized  R-F  signals  from  the  transient  recorder 
on  a  magnetic  disc,  and  later  to  read  the  stored 
records  and  process  them  as  desired.    In  practice, 
the  operator  manipulated  the  transducer  until 
the  beam  passed  through  the  desired  area,  using 
the  B-scan  image,  if  desired;  then  placed  the 
depth  marker  at  the  beginning  of  the  region  of 
interest,  checked  the  signal  on  the  oscilloscope, 
and  initiated  recording  with  the  pushbutton. 
This  was  found  to  be  quite  fast  and  convenient. 
The  system  is  depicted  schematically  in  figure  7. 

After  some  experimentation,  it  was  decided  to 
standardize  the  format  of  recorded  data  in  order 
to  simplify  the  signal  processing  programs. 
A  sample  rate  of  20  MHz  was  chosen,  giving  a 


B-SCAN 
SYSTEM 


TRONSOUCER 


PUL5ER 

— r~ 


DATA    ACQUISITION  PROCESSING 

a 

STORAGE 


OSCILLOSCOPE 


TRANSIENT 
RECORDER 


n  -J 

MANUAL 
ARMING 
CONTROL 


[—  MINICOMPUTER 


MAGNETIC 

DISC 


ANALOG  SIGNAL  PATH 
DIGITAL  SIGNAL  PATH 


Fig.  7.    A  schematic  diagram  of  the  system  used 
to  obtain,  store,  and  process  ultrasonic 
echoes . 


usable  signal  bandwidth  of  10  MHz.  Record 
lenngth  was  standardized  at  512  samples,  which 
represented  25.6  ys,  or  about  1.9  cm  of  tissue. 
This  was  found  to  be  long  enough  for  most  pur- 
poses, but  a  provision  was  made  to  store  four 
consecutive  records  if  desired,  representing 
7.7  cm  of  tissue.    Fourier  transforms  could  be 
calculated  on  512  point  records  in  10  seconds, 
and  cepstra  in  23  seconds,  fast  enough  for  inter- 
active studies  on  the  effects  of  various  types 
of  filtering. 

Two  types  of  experiments  have  been  run  to  col- 
lect data  for  various  purposes.    In  the  first 
type,  the  scanning  arm  of  the  B-scan  system  is  not 
used.    The  transducer  is  fixed  in  a  water  tank, 
suspended  from  a  micrometer  adjustable  frame. 
The  transducer's  beam  could  be  aligned  to  reflect 
from  the  water  surface  for  transducer  characteriza- 
tion and  reference  purposes,  on  targets  suspended 
from  a  rotatable  mount,  or  on  tissue  samples  pin- 
ned to  a  sponge  on  the  bottom  of  the  tank.    In  the 
second  mode,  the  scanning  arm  was  used  and  identi- 
fication of  tissues  was  made  using  the  images 
formed. 

Single  reflections  from  the  surface  of  the 
water  tank  were  used  to  obtain  data  on  the  effi- 
ciency and  spectral  characteristics  of  both  pur- 
chased and  fabricated  transducers,  and  to  verify 
the  structure  of  the  cepstrum  of  a  transducer  im- 
pulse response.    Impulse  responses  and  cepstra  for 
two  commercial  transducers  are  shown  in  figures  8 
and  9.    A  transducer  was  then  aimed  at  the  rotat- 
ing mount,  and  a  fixture  holding  three  1.1  mm 
diameter  rods  side  by  side  in  the  beam  at  a  spac- 
ing of  2.5  mm  was  attached,  as  shown  in  figure  10. 
The  array  of  rods  was  rotated  from  perpendicular 
to  the  beam  to  parallel,  and  on  to  perpendicular 
again,  and  the  reflections  recorded  at  five  degree 
increments  in  angle  to  provide  reflections  of 
various  spacings.    The  envelope  of  these  cepstra 
is  plotted  as  a  function  of  angle  in  figure  11. 
The  sinusoidal  dependence  of  spacing  on  angle  can 
be  clearly  seen.    The  short  spacing  is  seen  best 
on  one  side  of  90°,  and  the  long  spacing  on  the 


(a) 


(b) 


25.6 


TIME  .  MICROSECONDS 


Fig.  8. 


(a)  Impulse  response  of  a  commercial 
2.25  MHz  transducer. 

(b)  Impulse  response  of  a  commercial 
3.5  MHz  transducer. 


291 


12.8 


TIME  ,  MICROSECONDS 


Fig.  9.    (a),  (b)  Cepstra  of  the  impulse  responses 
of  8(a)  and  (b)  respectively. 


3.2  6.4  9.6 

TIME,  MICROSECONDS 


ANGLE  B 


Fig.  11.    Moduli  of  cepstra  recovered  from  the 
scheme  of  figure  10,  as  a  function  of 
angle. 


ROTATING 
MOUNT 


CYLINDRICAL 
RODS 


TRANSDUCER 


Fig.  10.    Schematic  diagram  of  a  scheme  to  produce 
three  identical  echoes  with  variable  spac- 
ing.   Three  1.1  mm  diameter  steel  rods  are 
mounted  2.5  mm  apart  center  to  center. 

other  because  a  slight  misalignment  of  the  rods 
caused  the  reflections  to  be  unequal  in  strength 
and  to  depend  on  angle.    At  angles  where  the 
separation  between  the  reflections  is  comparable 
to  the  pulse  width  (that  is,  about  .5  ys),  inter- 
ference can  be  seen  to  occur  between  the  peak  rep- 
resenting the  pulse  shape  and  that  representing 
the  periodicity. 

In  order  to  demonstrate  the  ability  of  the  cep- 
strum  to  detect  a  periodic  structure  in  biological 
tissue,  a  particularly  simple  sample  was  chosen: 
pig  liver.    Pig  liver  is  a  large  organ,  readily 
available  in  a  fairly  fresh  state,  and  having  a 
characteristic  and  simple  period  structure  [8]. 
Our  sample  was  several  centimeters  thick,  and  was 
observed  to  have  an  internal  structure  consisting 
largely  of  spherical  lobules  of  fairly  uniform 
size,  about  2  mm  in  diameter.    Several  echoes  were 
recorded  from  various  parts  of  the  liver,  using  a 
3.5  MHz  focused  transducer  in  the  water  tank.  The 
best  of  the  cepstra  recovered  is  shown  in  figure 


TIME  ,  MICROSECONDS 


Fig, 


12.    The  modulus  of  the  cepstrum  of  the  signal 
shown  in  figure  1 . 


12.    The  signal  from  which  it  was  derived  was 
shown  in  figure  1.    The  strong  peak  at  2.5  ys  cor- 
responds to  a  spacing  of  1.9  mm  between  reflectors, 
which  corresponds  well  with  the  observed  size  of 
the  lobules.    Other  cepstra  were  not  always  so 
simple. 

Cepstra  were  also  calculated  for  signals  from 
various  regions  of  human  livers  from  autopsies, 
as  well  as  from  thigh  muscles  and  the  area  of  the 
liver,  in  vivo.    The  details  of  the  structure  of 
both  the  power  spectrum  and  the  cepstrum  were 
found  to  vary  markedly  among  the  samples,  and 
even  between  two  consecutive  5  mm  long  regions  of 
single  records.    Some  cepstra  showed  substantial 
single  peaks;  some  did  not.    The  decay  rates  of 
the  cepstra  varied  substantially  also.    We  conclude 
that  the  tissues  studied  are  sufficiently  inhomo- 
geneous  to  prevent  easy  interpretation  of  these 
results.    A  broad  statistical  study  of  a  large  num- 
ber of  signals  documented  as  to  the  location  and 
possible  pathology  of  the  tissue,  original  trans- 
ducer response,  and  thickness  of  intervening  tis- 
sue, will  be  needed  to  prove  the  usefulness,  or 
lack  thereof,  of  such  techniques. 


292 


7.    Attenuation  Estimation 

Since  the  cepstrum  is  the  Fourier  transform  of 
the  log  power  spectrum,  low  pass  filtering,  or 
smoothing,  of  the  log  power  spectrum  may  be  done 
by  multiplying  the  cepstrum  by  a  window  function 
which  leaves  short  time  responses  untouched  and 
eliminates  long  time  values.    The  resultant  gated 
signal  may  then  be  Fourier  transformed  to  the 
frequency  domain  again.    If  the  variations  in  the 
spectrum  are  due  to  reflections  with  relatively 
large  spacing  tissue,  and,  hence,  are  rapid  com- 
pared to  the  finite  bandwidth  of  the  incident  beam 
of  sound,  the  rapidly  varying  spectral  components 
can  be  filtered  out,  leaving  a  good  approximation 
to  the  log  power  spectrum  of  the  incident  beam 
distorted  by  the  average  attenuation  in  the  tis- 
sue.   If  such  an  approximation  is  made  for  two 
consecutive  regions  of  tissue  along  the  same 
ultrasonic  beam  path,  the  difference  between  them 
should  exhibit  the  frequency  dependence  of  the  at- 
tenuation of  ultrasound  through  the  path  between 
them.    Such  a  calculation  does  not  require  as 
strong  an  assumption  about  the  variations  in  the 
log  power  spectrum  due  to  the  tissue  reflections. 
There  may  be  slow  variations;  if  they  are  the  same 
in  the  two  regions  of  tissue,  their  difference 
will  be  zero.    Since  slow  variations  in  the  log 
power  spectrum  arise  from  structure  fine  compared 
to  the  ultrasonic  pulse  length,  the  success  or 
failure  of  this  technique  to  predict  a  plausible 
frequency  dependence  for  the  attenuation  of  ultra- 
sound will  give  a  measure  of  the  similarity  of  the 
structure  of  the  two  regions  of  tissue  on  the 
scale  of  the  ultrasonic  pulse  length. 

Since  the  advantage  of  this  technique  would  be 
its  ability  to  measure  attenuation  in  vivo  non- 
invasively,  it  was  applied  to  the  data  recorded 


-+20r 


UJ 


o 
cl 


CD 

o 


(b) 


■201- 


FREQUENCY  (MHz) 

Fig.  13.    (a)  and  (b),  log  power  spectra  from  two 
consecutive  1  cm  long  regions  of  in  vivo 
human  liver  tissue  with  2  cm  center  spac- 
ing, recorded  with  a  2.25  MHz  transducer. 


1"  This  procedure  is  illustrated  in  figures 

13  and  14.    First  the  data  is  windowed  to  the  pro- 
per length  and  Fourier  transformed,  and  log  power 
and  spectra  obtained.    The  signal  to  noise  ratio 
has  been  artifically  decreased  to  20  dB  to  de-em- 
phasize the  random  spectral  components  outside  the 
useful  bandwidth  of  the  transducer,  1  to  3  MHz. 
The  cepstra  are  then  computed  and  low  pass  filtered 
with  a  sine-squared  window,  which  closely  approxi- 
mates a  Gaussian  window,  but  completely  cuts  off 
long  time  components.    Fourier  transforming  again 
yields  the  smoothed  log  power  spectrum.  Subtract- 
ing one  such  spectrum  from  the  other  and  dividing 
by  the  path  length  between  them  yields  a  curve 
which,  over  the  useful  bandwidth  of  the  transducer, 
estimates  the  frequency  dependence  of  the  loss  over 
that  path.  The  absolute  loss  is  not  estimated  here, 
although  with  attention  to  time-gain  control 
settings  and  assumption  of  equal  reflection  ampli- 
tudes in  the  two  regions,  it  could  be.    In  this 
example  in  which  the  signals  were  taken  from  two 
regions  about  1  cm  long  and  2  cm  apart  in  the  area 
of  the  liver  of  a  person  with  a  presumed  normal 
liver,  the  estimated  loss  follows  fairly  well  a 
linear  frequency  dependence,  although  with  a  coef- 
ficient of  0.8  dB/cm  MHz,  slightly  lower  than  the 
previously  reported  in  in  vitro  value  [9] 


20r 


UJ 


o 

Q. 


O 


T3 


< 
ZD 


111 

> 

< 


FREQUENCY  (MHz) 

Fig.  14.    (a)  Solid  line,  the  cepstrally  smoothed 
log  power  spectrum  derived  from  that 
of  figure  13(a).    Dashed  line,  the 
same  function  of  the  spectrum  of 
figure  13(b). 
(b)  Solid  line,  the  difference  between 
the  two  smoothed  spectra,  divided 
by  the  path  length.    Dashed  line, 
the  slope  of  the  attenuation  in  the 
frequency  range  where  both  spectra 
are  above  noise.    This  slope  is 
.8  dB/cm. 


293 


-+  20 


LU 


20 


0  


•20r 


(a) 


FREQUENCY  (MHz) 

Fig.  15.    Log  power  spectra  analogous  to  those  of 
figure  13,  but  from  a  different  region 
of  the  liver. 

An  interesting  phenomenon  has  been  found  in 
the  two  vn_  vivo  livers  and  the  one  i_n_  vitro  liver 
studied  with  this  technique  so  far.    In  many  of 
the  log  power  spectra  calculated,  there  is  a  deep 
notch,  representing  an  absence  of  reflection  at 
frequencies  varying  from  2.4  to  2.7  MHz.  This 
notch  has  been  seen  with  both  focused  and  unfo- 
cused transducers  of  both  2.25  and  3.5  MHz  center 
frequency.    This  phenomenon  violates  the  condi- 
tions under  which  one  can  get  a  good  estimate  of 
attenuation.    It  occurs  only  once  in  the  useful 
bandwidth  of  the  transducers  used,  and  so  cannot 
be  considered  rapidly  varying.    Its  frequency 
varies  enough  from  region  to  region  that  it  does 
not  cancel  when  two  spectra  are  subtracted.  The 
very  deep  notch  is  almost,  but  not  quite,  re- 
moved by  the  smoothing  operation.    Since  the 
attenuation  calculation  takes  the  difference  be- 
tween two  similar  functions,  it  is  sensitive  to 
the  small  error  which  remains.    An  example  of  this 
problem  is  shown  in  figures  15  and  16.    Such  a 
phenomenon  probably  arises  because  most  of  the 
reflections  in  a  region  are  coming  from  a  single 
type  of  structure:    one  with  a  constant  size  of 
a  fraction  of  a  millimeter.    Measurements  over  a 
broader  bandwidth  could  determine  more  exactly 
what  type  of  structure  this  is.    If  it  could  be 
identified,  there  might  be  clinically  useful  in- 
formation available. 

8.  Conclusions 

From  this  limited  study,  we  conclude  that: 
(a)    Cepstral  analysis,  by  converting  ultra- 
sonic signals  from  a  domain  in  which  the  output  is 
the  convolution  of  an  impulse  response  with  a  re- 
flection function  to  a  domain  in  which  it  is  a 
sum,  allows  the  separation  of  information  about 
the  distribution  of  spacings  between  reflectors 
from  information  about  the  incident  impulse. 


LU 

B 

a. 

o 
o 


20 


- 

/  \ 

\  \ 

\  / 

\  ^ 

\ 

1        1  I 

(a) 


o 


< 


UJ 

> 


UJ 

cr 


FREQUENCY  (MHz) 


Fig.  16.     (a)  Smoothed  log  power  spectra  from  those 
of  figure  15. 
(b)  Attenuation  estimated  from  figure 
16(a).    No  estimate  is  possible  in 
this  case. 

(b)    Two  potentially  useful  parameters  for  tis- 
sue characterization  in  vivo  are  the  form  of  the 
cepstrum  of  ultrasound  echoes  from  tissue,  which 
can  identify  periodicities  in  the  tissue;  and  at- 
tenuation coefficient  estimation  by  cepstral 
smoothing  of  log  power  spectra. 

A  lack  of  consistency  from  one  region  to 
another  within  a  tissue  suggests  considerable 
variability  in  the  tissue.    Further  work,  utiliz- 
ing statistical  studies  on  large  data  bases,  will 
be  required  to  verify  the  usefulness  of  these 
techniques.    Cepstral  analysis  should  prove  use- 
ful, since  averaging  over  similar  samples  will  en- 
hance the  tissue-dependent  information.    This  is 
in  marked  contrast  to  the  power  spectral  domain, 
where  averaging  over  similar  samples  tends  to 
eliminate  tissue-dependent  information  and  approxi- 
mate the  power  spectrum  of  the  measuring  system. 

Acknowledgments 

The  work  reported  in  this  paper  was  supported 
in  part  by  the  National  Science  Foundation  under 
Grant  No.  ENG  75-18681,  and  in  part  by  the  Kaiser 
Foundation. 

References 

[1]    Several  of  these  studies  are  reviewed  in: 

Dunn,  F.,  Ultrasonics  Attenuation,  Absorption, 
and  Velocity  in  Tissues  and  Organs;  and, 
Reid,  J.M.,  The  Scattering  of  Ultrasound  by 
Tissues,  in  Ultrasonic  Tissue  Characterization. 
M.  Linzer,  ed..  National  Bureau  of  Standards 
Spec.  Publ.  453,  pp.  21-28  and  29-47  (U.S. 
Government  Printing  Office,  Washington,  D.C., 
1976). 


294 


[2]    Bogert,  B.  P.,  Healy,  M.  J.  R.,  and  Tukey, 
J.  W.,  The  Frequency  Analysis  of  Time  Series 
for  Echoes:    Cepstrum,  Pseudo-Autocovariance, 
Cross-Cepstrum,  and  Saphe  Cracking,  in  Proc . 
Sym.  Time  Series  Analysis,  M.  Rosenblatt,  ed. , 
pp.  209-243  (John  Wiley  and  Sons,  Inc.,  New 
York,  1963). 

[3]    Stockham,  T.  C,  Jr.,  Restoration  of  Old 

Acoustic  Recordings  by  Means  of  Digital  Signal 
Processing,  Preprint,  41st  Convention,  Audio 
Engineering  Society,  New  York,  October  1971. 

[4]    Ulrych,  T.  J.,  Application  of  homomorphic 
deconvolution  to  seismology.  Geophysics  36 
(4),  650-660  (August  1971). 

[5]    Adaptive  Nonlinear  Signal  Processing  for 

Characterization  of  Ultrasonic  NDE  Waveforms, 
Task  2:    Measurement  of  Subsurface  Fatigue 
Crack  Size,  in  Technical  Report  AFML-TR-76-44, 
April  1976,  Section  6.4,  p.  54. 


[6]    Bracewell ,  R.,  The  Fourier  Transform  and  Its 
Application,  pp.  269-27MMcGraw-Hi  1 1 ,  New 
York,  1965). 

[7]    Oppenheim,  A.  V.  and  Schafer,  R.  W.,  Digital 
Signal  Processing  Chap.  10,  pp.  480-531 
TPrentice-Hall ,  Inc.,  Princeton,  1975). 

[8]    Lele,  P.  P.,  Mansfield,  A.  B.,  Murphy,  A.  I., 
Namery,  J.,  and  Senapati ,  N.,  Tissue  Charac- 
terization by  Ultrasonic  Frequency  Dependent 
Attenuation  and  Scattering,  in  Ultrasonic 
Tissue  Characterization,  M.  Linzer,  ed.. 
National  Bureau  of  Standards  Spec.  Publ.  453, 
pp.  167-196  (U.S.  Government  Printing  Office, 
Washington,  D.C.,  1976). 

[9]    Goldman,  D.  E.  and  Heuter,  T.  F.,  Tabular 
data  of  the  velocity  and  absorption  of  high 
frequency  sound  in  mammalian  tissues,  J_. 
Acoust.  Soc.  Am.  28,  35  (1956). 


295 


1 
t 


li 
ij 
If 
is 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


RECOGNITION  OF  PATTERNS  IN  ULTRASONIC  SECTIONAL  PICTURES 
OF  THE  PROSTATE  FOR  TUMOR  DIAGNOSIS 


W.  von  Seelen,^  A.  Gaca,^  E.  Loch,^  W.  Scheiding,^  and  G.  Wessels^ 

ilnstitut  f(Jr  Zoologie 
Arbei tsgruppe  III  (Biophysik) 
D-6500  Mainz,  Federal  Republic  of  Germany 

^Deutsche  Klinik  fUr  Diagnostik 
D-6200  Wiesbaden,  Federal  Republic  of  Germany 

^Battelle  Institut  e.V. 
D-600  Frankfurt/M. ,  Federal  Republic  of  Germany 


We  examined  the  prostate  with  ultrasonics  to  diagnose  tumorous  alterations  of  the 
organ.    We  scanned  directly  from  the  abdominal  wall  through  the  filled  bladder  (trans- 
vesical).   The  research  is  aimed  at:    1)  improvement  and  mathematical  preparation  of 
the  ultrasonic  pictures  to  assist  the  physician  in  his  diagnosis;  and  2)  ascertainment 
of  significant  parameters  which  allow  the  classes  "normal  prostate,"  "adenoma"  and 
"carcinoma"  to  be  distinguished  in  screening  examinations.    The  results  show  that  with 
ultrasonics,  adenomas  and  carcinomas  are  recognizable  in  96  percent  of  the  examined 
patients  and  can  be  differentiated  from  normal  prostate  tissue.    The  palpation  finding 
was  used  as  a  reference  in  most  cases. 


Key  words:    Pattern  recognition;  prostate  tumors;  ultrasound. 


1.  Introduction 

Carcinoma  of  the  prostate  ranges  third  in  cancer 
mortality  in  the  Federal  Republic  of  Germany.  Be- 
sides the  subjective  rectal  palpation  findings  which 
give  the  physician  signs  of  organic  alterations,  it 
is  useful  to  look  for  other  examination  methods. 

The  aim  of  this  investigation  is  to  analyze  the 
value  of  the  ultrasonic  scan  for  differential  diag- 
nosis and  treatment  control  of  tumor  diseases  and 
to  check  in  establishing  preclinic  findings  the 
possibility  of  a  screening  procedure  allowing  a 
guick  separation  between  suspect  and  non-suspect 
cases  in  large  groups  of  patients. 

The  apparatus  available  for  the  methodical  part 
of  the  investigations  are  two  ultrasonic  scanners: 
Combison  II  Kretz  (Compound-Scanner)  Vidoson,  Sie- 
mens (Realtime-Scanner)  as  well  as  a  computer  (PDP 
11/34).    The  apparatus  is  shown  schematically  in 
figure  1.    This  study  includes  both  examinations  in 
vivo  and  some  examinations  in  vitro.  Transvesical 
representation  of  the  prostate  was  chosen  for  in 
vivo  examination.    The  ultrasonic  pictures  are 
taken  from  above  the  organ  by  scanning  in  sections 
of  3  mm;  the  angle  at  which  the  sound  waves  enter 
is  either  «  =  15°  or  a  =  20°  from  the  vertical. 

Figure  2  shows  the  position  of  the  organ  as  well 
as  the  direction  of  entry  of  the  sound  used  to  pro- 
duce the  pictures  discussed  below.    The  freguencies 
are  2.25  MHz  and  4  MHz.    Linear  scanning  is  used 
to  cover  the  area  of  the  organ  with  a  slightly  al- 
tered angle  of  entry  for  the  forward  and  reverse 


runs.    The  B  pictures  gained  in  this  way  are  re- 
corded on  a  video  tape  which  serves  as  mass  storage 
For  the  computer  analysis  the  video  signal  can  be 
digitized  by  a  specially  developed  interface  and 
read  into  a  512  x  512  matrix;  a  region  of  interest 
can  be  marked  out  within  the  picture.    In  addition, 
it  is  possible  to  feed  the  non-demodulated  HF  sig- 
nal or  the  A  signal  into  the  PDP  11  computer.  Two 
control  techniques  were  employed  in  the  analysis  of 
the  ultrasonic  pictures:    comparison  with  the  urolo 
gists'  palpation  findings  and  in  vitro  examinations 
of  dissection  material  taken  during  operations. 

The  ultrasonic  pictures  reproduce  the  geometri- 
cal arrangement  and  form  of  the  density  gradients 
in  the  tissue  which  are  situated  perpendicular  to 
the  direction  of  the  sound.    Since  not  enough  is 
known  about  the  reflecting  properties  of  the  vari- 
ous tissues,  the  picture  features  for  the  different 
picture  classes  cannot  be  defined  exactly  a  priori . 
The  study  was  therefore  carried  out  in  the  follow- 
ing stages: 

1.  Definition  of  a  limited  learning  sample 
which  can  be  diagnosed  as  certainly  as  possible 
with  the  palpation  findings  and  histological 
control s . 

2.  Determination  of  as  many  apparently  signifi- 
cant features  as  possible. 

3.  Generation  of  an  adaptive  classifier  which 
begins  with  a  fixed  initial  weighting  of  the 
features.    This  weighting  is  changed  in  the  course 
of  the  classification  of  the  learning  sample  so  as 
to  eliminate  errors.     In  this  way  the  physician 


297 


Fig.  3.    Ultrasonic  picture  of  a  normal  prostate. 


Fig.  4.    Ultrasonic  picture  of  an  adenomatous 
prostate. 


Ultrasonic  picture  of  a  carcinomatous 
prostate. 


supervises  the  determination  of  features  and  class 
1 imits. 

4.    Ascertainment  of  the  validity  of  the  indi- 
vidual features  to  determine  a  precise  set  of 
distinguishing  features  which  does  not  involve  too 
difficult  or  costly  application.    The  criterion  for 
validity  is  provided  by  the  alteration  in  detection 
rate  for  the  relevant  feature. 

Figures  3,  4  and  5  each  show  one  representative 
of  the  classes  normal,  adenomatous  and  carcinomatous 
respectively.    While  the  normal  organ's  interior  is 
characteristically  largely  echo  free,  the  adenoma 
appears  highly  structured.    The  carcinoma  in  the 
right  half  causes  a  "shadow"  which  penetrates  the 
capsule  areas.    Other  features  of  the  pictures  will 
be  discussed  below. 

2.    Preprocessing  of  the  Pictures 

Preprocessing  the  pictures  is  intended  to  allow 
the  doctor  to  make  the  most  certain  possible  diag- 
nosis.   For  this  it  is  necessary  to  eliminate  dis- 
turbances and  distortions  in  the  picture  and  to 
create  the  possibility  of  accentuating  features  re- 
levant to  suspected  cases,  if  this  is  required.  To 
this  end  the  operations  described  below  were  im- 


298 


plemented  on  the  digital  computer  and  then  first 
tested  by  subjecting  them  to  the  same  classifica- 
tion procedure  as  the  original  pictures.    The  fol- 
lowing operations  were  carried  out  on  the  pictures 
(for  different  numbers  of  patients  in  each  case). 

a.  Definition  of  a  "region  of  interest."  This 
operation  consists  in  limiting  the  organ  area  on 
the  screen  and  reducing  the  amount  of  data  taken 
over  in  the  computer  from  the  picture  to  128  x  128 
picture  points  (8  bits  per  point)  which  thus  refer 
to  the  region  of  interest.    This  limitation  of  the 
picture  is  carried  out  by  the  physician  treating 
the  case. 

b.  Standardizing  the  intensity  of  the  picture. 
The  space  dependent  intensity  of  the  picture  x(r,s) 
in  the  B  scan  picture  varies  a  great  deal  from 
patient  to  patient  (age,  corpulence,  etc.)  and  also 
depends  on  the  amplification  of  the  experimental 
device.    Therefore,  parameters  dependent  on  in- 
tensity should  only  be  used  in  the  diagnosis  when 
x(r,s)  is  divided  by  the  mean  value  x(r ,s )  deter- 
mined in  a  defined  organ  area.    The  resulting  pic- 
ture is  the  basis  for  several  computer  operations. 

c.  Inverse  filtering.    To  eliminate  the  loss 
in  lateral  focus  due  to  the  form  H, (r,s)  of  the 
sound  beam  as  far  as  possible,  the  values  ascer- 
tained under  water  for  this  space  dependent  cou- 
pling were  used  to  determine  the  filter  charac- 
teristic value  H.(r,s).    The  conditions  for  the 
realization  of     (r ,s) 


F(H|^(r,s))  •  F(H.(r,s))  =  1 


were  replaced  by 


E(H|^(r,s)) 


=  1 


1  +  F(H.(r,s)) 


(b) 


for  numerical  reasons.    F  characterizes  the  Fourier 
transform  according  to  the  two  space  coordinates 
r,s.    Figure  5a  shows  the  sectional  picture  of  a 
0.5  mm  thick  wire  in  the  tank  of  water  as  well  as 
its  filtered  representation  with  improved  radial 
focus . 

d.    Symmetrical  filtering.    The  convolution  of 
the  picture  X(r,s)  with  any  optional  symmetrical, 
and  thus  phase  free  transmission  function  H(r,s) 
allows  a  far-reaching  alteration  of  the  pictures, 
when  one  varies  the  parameters  nix,  ni2>  ^i'  ^2  in 


Y(r,s)  =  X(r,s)  *  H(r,s] 


with 


H(r,s)  =  Um^e 


Two  phase  free  band  pass  filters  with  different 
mean  frequencies  Umj  and  Um2  were  realized  for 
miBi  =  m2B2  and  B2  >  B^.    Figures  6b  (F)  and  5c  (G) 
show  the  result  of  the  operation  for  Um2  >  Umj.  The 
classes  can  be  better  distinguished  in  the  filtered 
version  of  the  picture  (F,  Umj)  than  in  the  original 
picture  (cf.  Section  4).    The  second  filter  opera- 
tion (G)  is  interpreted  so  that  the  picture  ampli- 
tude is  approximately  proportional  to  the  gradient 
of  the  intensity  in  the  original  picture.  This 
variable  has  to  be  taken  into  consideration  at  the 
boundary  of  carcinomas. 


(c) 

Fig.  6.    Sectional  picture  of  a  wire  before  and 
after  inverse  filtering  (a);  band  pass 
filtered  sectional  picture  of  a  carcin- 
omatous organ  with  Umi  (F)  (b);  and  Uma 
(G)  (c). 


299 


e.    Phase  dependent  filters.    It  is  easier  to 
judge  the  pictures,  if  the  function  H(r,s)  is  not 
symmetric,  i.e.: 

H(r,s)  =  Umie-(^'"^')/Bf .  ^„^,-i(r-r,)^ns-s,)^)/4 

with  Bf<<  Bj.    In  this  case  the  Fourier  transform 
F  (H(r,s))  is  complex  and  a  phase  shift  exists.  In 
this  way  a  picture  is  produced  which  is  interpreted 
as  pseudo  three-dimensional  when  viewed.    If  H(r,s) 
characterizes  a  differentiating  space  filter  com- 
bined with  a  nonlinear  characteristic  curve,  then 
the  same  effect  occurs  together  with  an  accentuation 
of  the  intensity  modulation.    Figure  7  shows  two 
examples  for  different  H(r,s)  (paper  in  preparation). 
All  filter  operations,  insofar  as  they  are  linear, 
can  be  combined  to  one  ooeration. 


Fig.  7.    Filtering  with  asymmetrical  functions 
H(r,s). 

f.    Equidensities .    It  is  easier  to  analyze  the 
weak  modulations  in  the  area  of  the  organ  more 
exactly  when  the  low  pass  filtered  original  pic- 
ture is  provided  with  lines  of  equal  intensity. 
Figure  8  shows  an  example  with  an  increased  number 
of  picture  points. 


Fig.  8.    Carcinomatous  prostate  represented  in 
equidensities  in  a  low  pass  filtered 
sectional  picture. 

It  depends  on  the  problem  in  question  which  of 
the  procedures  described  is  best  suited  for  the 
physician's  diagnosis.  Only  the  F  and  G  pictures 
have  been  tested  with  the  classification  procedure 
(cf .  Section  4) . 

3.    Feature  Extraction 

The  problem  of  feature  extraction  has  to  be  solv 
ed  in  two  ways.    The  reference  vectors  are  deter- 
mined on  the  basis  of  the  physician's  palpation 
findings  and  the  vectors  x-  ai"e  obtained  from  the 
ultrasonic  pictures.    The  palpation  finding  takes 
into  consideration  size,  consistency,  surface  state 
and  simple  form  parameters  [1-5]^.  The  analysis  of 
the  ultrasonic  picture  refers  to  global  and  local 
features  which  in  the  first  phase  are  still  partly 
obtained  by  visual  scrutiny  of  the  ^reen  picture. 
Should  these  features  prove  to  be  effective  in  dis- 
tinguishing classes,  the  procedure  will  be  complete 
ly  automated. 

In  particular,  the  following  features  are  quanti 
fied: 

1.  Longitudinal  diameter  of  the  organ. 

2.  Rise  in  the  base  of  the  bladder. 

3.  Local  gaps  in  the  capsule. 

4.  Coarse  parallel  fibrous  structure  in  the 
interior  of  the  organ. 

5.  Fine  fibrous  structure  in  the  interior. 
The  automatic  determination  of  features  4  and  5 

by  multiple  correlation  of  the  picture  with  texture 
pictures  in  which  the  form  of  the  elements  is  op- 
tional (e.g.,  elliptical)  but  their  location  is 
statistical,  is  being  implemented  at  present. 

After  the  pictur'e  has  been  transferred  from  the 
video  recorder  to  the  computer,  the  following  param 
eters  are  extracted  according  to  the  definition  of 
the  region  of  interest: 

6.  Surface  of  the  sectional  picture  Fg. 

7.  Autocorrelation  (^^^(O'O) 

8.  The  standardized  signal  power  in  the  region 
of  interest,  i.e.,  <j)^^(o,o)/x(r,s)2. 

^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


300 


9.    The  amplitude  density  distribution  p(x(r,s)) 
with  the  relevant  first  three  moments  or 
central  moments  respectively 

E(x(r,s)),     E(x(r,s)  -  x(r,s))2, 

E(x(r,s)  -  x(r,s))3. 

10.  Curve  of  the  autocorrelation  function  along 
a  line  through  the  suspect  zones  (j)  (r'). 

^r^r 

11.  Coherence  width  and  relative  extremes  of  the 
autocorrelation  function  <t)xx('^)  along  a  line 
through  suspect  zones. 

12.  Power  and  energy  above  a  threshold  or  with- 
in an  amplitude  window  respectively.  This 
operation  has  so  far  only  been  implemented 
on  60  statistically  selected  patients. 

In  order  to  include  local  form  parameters  in  the 
consideration,  the  features  6  through  11  were  de- 
termined in  the  original  picture  and  in  two  filter- 
ed pictures  (F,G)  and  classified  separately.  In 
this  way  the  efficiency  of  the  filtering  in  dis- 
tinguishing these  features  was  tested.    The  effect 
of  picture  standardization  and  parameter  12  have 
so  far  only  been  studied  on  60  patients.    The  va- 
idity  of  all  the  parameters  mentioned  is  being 
tested  with  an  adaptive  classifier. 


4.  Classification 

If      is  a  feature  vector,  then  classification 
consists  in  the  division  of  feature  space  by  class 
limits  so  that 

1  Dj(^)  i,j  =  1,2,  ...  m,  i  M 

when       is  correctly  classified;  D  is  a  discriminat- 
ing function  [6].    In  this  project  an  adaptive 
linear  classifier  was  chosen  at  first  with 


Di(Zi)  =  E     +  ""^w. 

k=l  ^ 

The  variable    W-j  designates  the  kth  component  of 
the  reference  vector  W'l  which  is  generated  in  p 
steps  with  the  aid  of  a  learning  sample  according 
to  the  rule 

-i,6+i  ^  -i  6     '^'^  ^i        correctly  classified 
and 

-i,6+i  "  -i,6  ^  "^6  *  ^i,6  ''^  i^i  incorrectly 
classified  with 


In  the  case  of  a  Gaussian  distortion  of  the 
features,  the  decision  rule  minimizes  the  quadratic 
distance  between  the  reference  vector  and  the  vector 
to  be  classified.    The  decisions  on  a  correct  or  in- 


correct classification  in  the  adaptive  process  de- 
scribed above  are  based  on  the  physician's  palpa- 
tion findings.    After  generation  of  the  reference 
vector,  the  pictures  were  classified  in  stages. 
First,  the  set  of  features  1  through  5  was  applied 
(Part  I),  then  the  parameters  6  through  11  for  the 
original  pictures  as  well  as  two  filtered  versions 
of  the  picture  (Part  II)  and  then  parameter  12  on 
intensity  standardized  pictures  for  60  patients. 
A  classification  on  the  basis  of  the  complete 
featureset  (I  +  II  +  III)  will  be  realized  if 
feature  12  is  determined  for  all  patients.  The 
class-specific  reference-vectors  were  established 
with  25  patients  per  class  in  the  first  stage  of 
the  project.    Therefore  the  following  number  of 
patients  can  be  considered  as  a  test  data  set. 

Part  I.    After  taking  palpation  findings  from 
500  patients,  the  parameters  1  through  5  (Section 
3)  were  extracted  and  the  classes  normal  (N)  and 
suspect  (adenomatous  (A),  carcinomous  (C))  distin- 
guished.   The  physician's  diagnoses  were: 

a)  97  patients  normal 

b)  324  patients  adenomatous  or  carcinomatous 

c)  79  patients  nonspecif ical ly  suspect  (prosta- 
titis, congestion  etc.). 

If  P(N/A  +  c)  characterizes  the  probability  that 
a  patient  classed  as  normal  nevertheless  has  an 
adenoma  or  carcinoma  (false  negative)  one  obtains 
the  following  results  for  false  classification.  In 
the  case  of  unequivocal  diagnosis: 

P(N/A  +  C)  =  3.2  %  (false  negative) 
P(A  +  C/N)  =  7.4  %  (false  positive) 

If  the  nonspecific  suspect  cases  are  also  taken 
into  consideration  then  P(N/A  +  C)  =  7.2  percent. 
The  increase  in  error  due  to  group  c  can  be  further 
lowered  with  more  precise  examination.    If  the  total 
number  of  false  classifications  is  considered,  the 
probability  of  error  is  3.9  percent.    In  the  3  group 
problem  the  mean  error  rate  increases  to  8  percent. 

Part  II.    The  classification  of  the  pictures 
from  198  patients  according  to  the  features  6 
through  11  resulted  in  a  mean  error  of  11.9  per- 
cent for  the  2  group  problem  (normal /not  normal) 
with  the  original  pictures  and  10  percent  for  the 
filtered  version  F  and  14  percent  for  the  filtered 
version  G.    First  results  are  now  available  for 
studying  the  weighting  of  the  individual  features 
and  altered  decision  strategies  but  the  analysis 
has  not  been  completed. 

Part  III.    Using  feature  12  presupposes  stand- 
ardizing the  picture  intensity.    In  a  set  of  60 
patients  selected  statistically  the  rate  of  error 
for  the  2  group  problem  was  0  percent.    The  mean 
error  for  the  3  group  problem  was  7  percent  and 
P(C/N  +  A)  =  0  percent.    This  study  is  at  present 
being  extended  to  the  whole  group  of  patients  (500). 
An  improvement  in  the  classification  can  be  expected. 

As  the  features  in  the  three  classification  ex- 
periments are  at  least  partially  independent  from 
each  other,  it  is  to  be  expected  that  the  rate  of 
error  can  be  reduced  with  a  combined  set  of  fea- 
tures.   The  systematic  study  of  these  combinations, 
which  are  intended  to  attain  a  better  feature  vector 
with  a  few  easily  calculable  components  has  not  yet 
been  completed. 

The  further  studies  are  concerned  with  ascer- 
taining a  histologically  tested  learning  sample, 
determining  the  sharpest  distinguishing  parameter 
and  obtaining  a  classifier  which  takes  the  statis- 
tical parameters  of  the  features  into  consideration. 


301 


References 


[1]    Gaca,  A.,  Loch,  t.  G.,  Scheiding,  U.,  von 

Seelen,  W.  ,  and  Wessels,  G. ,  Ultraschallunter- 
suchungen  der  Prostata  zur  Erkennung  von  Tumor- 
erkrankungen.    Report  BF-R-62. 991-3  (1977) 
Bundesministerium  fUr  Forschung,  Bonn,  Kennedy- 
Allee.    (This  report  contains  a  comprehensive 
list  of  the  relevant  literature.) 

[2]    King,  W.  ,  Kieineyer,  W.  ,  Mark,  R.  ,  Boyce,  W.  H., 
and  McKinney,  W.  M. ,  Current  status  of  prosta- 
tic echography,  J.  Amer.  Med.  Ass.  226  (4), 
(1973). 

[3]    Takahashi ,  H.  and  Ouchi ,  T.,  Ultrasonic  diag- 
nosis in  the  field  of  urology.    First  Report, 
Japanese  Medical  Ultrasonics,  pp.  7-10, 
Tokyo,  (1963). 

[4]    Takahashi,  H.,  and  Ouchi,  T.,  Ultrasonic  diag- 
nosis in  the  field  of  urology.    Second  Report, 
Japanese  Medical  Ultrasonics,  pp.  35-37,  Tokyo, 
T1964). 

[5]    Watanabe,  H.,  Igari,  D. ,  Tanahashi ,  Y.,  Hasada, 
K. ,  and  Saitoh,  M. ,  Diagnostic  application  of 
ultrasonography  to  the  prostate.  Invest.  Urol. 
8,  548-559  (1971). 


[6]    Fu,  K.  S.,  Digital  Pattern  Recognition 
(Springer-Verlag,  Berlin  1976) . 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


RECENT  DEVELOPMENTS  IN  OBTAINING  HISTOPATHOLOGICAL  INFORMATION 
FROM  ULTRASOUND  TISSUE  SIGNATURES 


K.  Preston,  Jr.,^  M.  J.  Czerwinski,^  M.  L.  Skolnik,^  and  D.  E.  Leb^ 

^Department  of  Electrical  Engineering 

Carnegie-Mellon  University 
Pittsburgh,  Pennsylvania    15213,  U.S.A. 

^Ultrasound  Laboratory 
University  of  Pittsburgh  Medical  Center 
Pittsburgh, Pennsylvania    15213,  U.S.A. 

^Department  of  Medicine 
University  of  Pittsburgh 
Pittsburgh, Pennsylvania    15261  U.S.A. 


Statistical,  signal -analytic  techniques  may  be  applied  to  ultrasonic  pulse-echoes 
received  from  organs  of  the  human  body.    Simultaneous  tissue  biopsies  of  these  organs 
may  be  sectioned,  stained,  and  measured  by  the  computerized  optical  microscope  using 
various  image-analytic  techniques.    This  paper  reports  some  preliminary  work  along 
these  lines  as  related  to  the  kidney  and  liver. 


Keywords:    Picture  processing;  signal  analysis;  ultrasound. 


1.  Introduction 

In  the  1950s  Wild  and  Reid        noticed  that 
there  appeared  to  be  a  correlation  between  the 
properties  of  the  A-scan  in  mammography  and  the 
histopathology  of  breast  tissue  in  human  subjects. 
Later  Fry  [2]  compared  the  histopathology  of  ani- 
mal tissue  (feline  and  porcine)  with  the  B-scan 
echogram.    More  recently  Mountford  and  Wells  [3], 
using  digitized  A-scans  from  both  normal  and 
cirrhotic  subjects,  found  that  specific  signal 
parameters  (amplitude,  rise  time,  etc.)  corre- 
lated with  liver  pathology.    Ossoinig  [4], 
Fields  et  al.  [5],  and  Taylor  and  Milan  [6]  made 
further  observations  of  B-scan  texture,  amplitudes 
and  positions  of  the  peaks  in  the  demodulated  A- 
scan,  and  A-scan  amplitude  probability  distribu- 
tion functions  ("histograms"),  respectively. 

The  advent  of  high-speed  digitizers  and  inex- 
pensive digital  computers  has  now  opened  this  field 
to  many  new  investigations.    One  of  these  investiga- 
tions was  commenced  by  the  authors  under  a  grant 
from  the  National  Science  Foundation  (APR75-08154) 
in  mid-1975.    A  preliminary  report  was  presented  by 
Preston[7]  at  the  first  meeting  of  this  internation- 
al seminar  at  Gai thersburgh  in  1975.    The  Carnegie- 
Mellon  University  program  is  being  conducted  along 
the  lines  of  a  project  in  classical  pattern  recog- 
nition in  distinction  to  a  project  in  biological 
modeling.    The  work  is  being  conducted  on  both 
humans  and  animals  in  vivo  in  cooperation  with  the 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


Departments  of  Radiology,  Penology,  and  Pathology 
of  the  University  of  Pittsburgh  Health  Center. 
Each  subject,  or  more  specifically,  the  target  organ 
within  the  subject,  is  considered  to  reside  in  Sig- 
nal Space.    This  space  is  interrogated  in  three 
ways.    A  standard  B-scan  is  made  using  commercial 
equipment  (Picker  Echoview)  and,  at  the  same  time, 
several  A-scans  are  digitized  using  a  special- 
purpose,  digitizing  system  (Stahl  Research  Labs.). 
Finally,  in  addition  to  the  ultrasonogram  and  the 
digitally  recorded  A-scan,  a  tissue  biopsy  is  made 
of  the  organ  under  examination.    These  interroga- 
tions furnish  information  which  constitutes  what  we 
term  "Data  Space. " 

A  visual  diagnosis  is  made  from  the  ultrasono- 
gram by  the  staff  of  the  Ultrasound  Laboratory  and 
is  combined  with  the  histopathological  diagnosis 
carried  out  by  the  Department  of  Pathology.  This 
furnishes  information  on  disease  type  and  severity. 
At  the  same  time  a  numerical  analysis  is  performed 
on  the  digitized  A-scan  using  the  statistical  meth- 
ods described  below.    These  various  methods  of 
analysis  of  the  information  in  Data  Space  provide 
us  with  numerical  information  which  is  contained  in 
Measurement  Space.    These  measurements,  taken  over 
a  population  of  patients,  are  then  processed  in  a 
supervised  fashion  (correlation  of  measurements 
with  predetermined  disease  types)  using  a  standard 
discriminant  function  software  package  (University 
of  California  BMD07M)  to  generate  points  in  Classi- 
fication Space.    Finally,  images  of  the  tissue  biop- 
sy are  digitized  by  the  Biomedical  Image  Processing 
Group  at  Jet  Propulsion  Laboratory  and  analyzed 
using  the  Biomedical  Image  Processing  Unit  of  the 
Department  of  Radiation  Health  at  the  University  of 


303 


Pittsburgh.    The  balance  of  this  paper  discusses 
A-scan  digitizations,  A-scan  analysis,  image  digiti- 
zation and  processing,  and  reports  results  to  date 
in  these  areas. 

2.    Materials  and  Methods 

The  ultrasound  digitizer  is  the  basic  unit  used 
for  digitizing  A-scans.    The  digitizer  (Stahl  Re- 
search Labs,  model  4036)  is  a  complete  digitizing 
system  rather  than  a  simple  analog/digital  con- 
verter.   Its  block  diagram  is  shown  in  figure  1. 
The  unit  requires  only  two  inputs,  namely  a  syn- 
chronizing or  gating  signal  and  the  analog  wave- 
form to  be  digitized  and  stored.    The  digitizer 
runs  under  control  of  an  internal  15  megahertz  clock. 
Burst  length  and  skip  burst  values  are  entered  us- 
ing digi switches  on  the  control  panel  of  the  model 
4036.    When  a  synchronizing  pulse  is  received,  the 
digitizer  proceeds  to  sample  and  digitize  the  input 
data  at  a  15  megahertz  rate  until  the  number  of 
words  stored  equals  the  burst  length.    The  unit  then 
begins  counting  additional  synchronizing  pulses  un- 
til the  count  has  reached  the  number  set  for  the  skip 
burst.    When  this  count  has  been  reached,  the  next 
synchronizing  pulse  causes  another  burst  to  be  digi- 
tized.   This  continues  until  the  4096  word  memory  is 
full  (each  word  is  one  8-bit  byte).    At  the  present 
time  we  are  using  a  burst  length  of  128  which  cor- 
responds to  8.5  microseconds  of  recorded  signal 
over  which  the  pulse-echoes  are  received  from  about 
0.65  centimeter  of  tissue. 

The  digitizer  incorporates  a  modem  and  complete 
memory  teletype  (Texas  Instruments  model  700)  having 
twin  digital  cassettes  each  of  which  is  capable  of 
storing  250,000  words  of  digitized  A-scan  data. 
When  the  user  desires  to  transmit  data  to  a  remote 
computer,  the  modem  is  interfaced  via  the  telephone 
system,  a  dial-up  connection  is  made,  and  the  digi- 
tal cassettes  are  activated  and  transfer  their  data 
via  the  communication  link.    In  our  present  program 


the  computer  used  is  a  Digital  Equipment  Corpora- 
tion PDPIO  model  KA  at  Carnegie-Mellon  University. 

In  all  experiments  to  date,  the  A-scan  is  digi- 
tized directly  from  the  output  of  the  radio  frequen- 
cy (rf)  amplifier.    The  unit  employed  for  both  puls- 
ing the  transducer  and  amplifying  the  A-scan  is  a 
Panametrics  model  5055  custom-modified  to  provide 
bandwidth  to  20  MHz.    In  our  experiments  we  have 
used  the  model  5055  at  its  full  gain  setting  (60  dB) 
and  with  full  bandwidth  with  the  interval  damping 
resistor  set  to  250  ohms.    The  organs  of  interest 
have  been  the  liver  and  kidney  for  which  we  obtain 
signal  levels  at  the  output  of  the  model  5055  rf 
amplifier  ranging  from  a  few  tens  to  a  few  hundreds 
of  millivolts.    (The  least  significant  digit  of  the 
digitizer  corresponds  to  approximately  8  millivolts.) 

The  transducer  utilized  is  a  Panametrics  custom- 
made  biopsy  transducer  whose  characteristics  have 
been  measured  by  the  Acoustics  Branch,  National 
Bureau  of  Standards,  and  are  presented  in  figure  2. 
As  can  be  seen,  the  resonant  frequency  is  close  to 
4  MHz  with  a  25  percent  3  dB  bandwidth. 


Fig.  2.    Transducer  1-way  frequency  response  as 
determined  by  the  National  Bureau  of 
Standards . 


MODE  CONTROL 


EXTERNAL  TRIGGER 


EXTERNAL  SYNC 


ANALOG  DATA 


BURST 


WORD 


AUDIT 


15  MHz  CLOCK 


SKIP  BURST  CONTROL 
0,32,  ■••  ,  512 


BURST  LENGTH 
1  28,256,  ••■4096 


BURSTS  PER 
BLOCK 


RECYCLE  RATE 


MEMORY  UPDATE  (FIXED  RECYCLE) 


PITCH 
CONTROL 


DIGITAL/ANALOG 
CONVERTER 


UART 


SINGLE  SIDEBAND 
MODULATOR 


TI700 
TELETYPE 


SLOW  AGC 


MODEM 


4096  WORD 
HIGH-SPEED 
MEMORY 


TWIN 
DIGITAL 
CASSETTES 


EAPHONES 


DIGITAL  COMMUNICATIONS  CHANNEL 


Fig.  1.    Block  diagram  showing  the  three  modes  of  operation  of  the  Stahl  Research 
Laboratory  model  4036  digitizer. 


304 


SAMPl (     IHC(  5  J 


Fig.  3.    Five  signal  regimes  for  the  normal  liver: 
(a)  normalized  rf  waveform;  (b)  rectified; 
(c)  video;  (d)  frequency  spectrum;  (e) 
deconvolved  spectrum.    Here  time  samples 
are  taken  each  100  ns  with  the  frequency 
domain  covering  0  to  5  MHz  in  0.08  MHz 
intervals. 

3.    A-Scan  Analysis 

Five  signal  regimes  (see  figs.  3  and  4)  are  used 
in  the  analysis  of  the  waveform:    (1)    Direct  rf 
waveform,  (2)  Rectified  rf  waveform,  (3)  Video  wave- 
form (the  rf  envelope),  (4)  The  rf  Wiener  spectrum, 
(5)  The  selectively  deconvolved  rf  Wiener  spectrum. 
For  regime  (5)  the  frequency  spectrum  of  the  trans- 
mitted signal  is  used  to  normalize  the  spectrum  of 
the  received  A-scan,  except  at  those  frequencies 
where  the  uncertainty  in  determining  the  transmitted 
spectrum  is  large.    Computationally,  the  selectively 
deconvolved  spectrum  D(a))  is  given  by 

S.(co)  /      1     \  /  \ 

D(co)  =      +  S .  (t.)   

S  fo))  \l  +  c2/        '       \1  +  c2/ 


Fig.  4.    The  five  signal  regimes  in  the  same  order 
as  in  figure  7  for  the  normal  kidney. 

Figure  5  shows  a  computer  generated  graph  of  four 
sequential  pulse  echoes  from  a  phantom  consisting 
of  a  multiplicity  of  polyethylene  disks  immersed  in 
water.    This  figure  shows  the  graphical  plot  (above) 
and  (below)  a  graytone  M-scan  rendition.    The  slight 
shift  in  pulse  position  which  is  evident  in  the  M- 
scan  output  is  an  artifact  due  to  synchronization 
problems.    The  pulse  consists  of  about  four  cycles 
at  the  resonant  frequency  of  the  transducer  and  is, 
therefore,  about  one  microsecond  in  duration.  Fig- 
ure 6  shows  the  Wiener  spectra  of  the  same  four 
pulses  presented  in  figure  5. 


Fig.  5.    Digitally  produced  A-scan  plots  and  M-mode 
graphs  of  pulse-echoes  reflected  from  a 
plane  surface. 


where  S^-(a))  is  the  spectrum  of  the  direct  rf  signal, 
Sp(u)  is  the  spectrum  of  the  transmitted  signal  from 
which  the  A-scan  echoes  are  produced,  and  quantity 
c  is  the  coefficient  of  variation  of  the  transmitted 
pulsed  (taken  over  an  ensemble  of  transmitted 
(pulses).    This  quantity  is  in  itself  a  function 
of  the  frequency  o). 


Fig.  6.    The  frequency  spectra  corresponding  to  the 
pulse-echoes  shown  in  figure  9. 


305 


Using  a  burst  length  of  128  words,  the  full 
memory  holds  32  bursts.    We  are  currently  employ- 
ing two  settings  of  skip  burst,  namely,  0  and  256. 
At  a  skip  burst  setting  of  0  the  entire  recording 
takes  place  during  32  pulses  which,  with  the  Pick- 
er Echoview  running  at  a  PRF  of  1  kHz,  takes  32 
milliseconds.    The  skip  burst  setting  of  256 
lengthens  the  recording  interval  to  8  seconds. 
A  comparison  of  recordings  taken  at  the  two  skip 
burst  settings  provides  a  measure  of  signal  sta- 
tionarity  (see  figs.  7  through  14). 


\y\A^/^ —  --w\/^^^ — 

\/\/\/\ —  \I^J\/^/^^^ — ^ — — — -^.'v^ 


Fig.  7.    Thirty-two  pulse-echoes  obtained  from  a 

normal  region  of  the  human  kidney  in  vivo. 
Echo  separation  is  a  0.001  second. 

Observation  of  figures  12  and  14,  which  were 
taken  at  the  skip  burst  256  setting,  indicate  that 
long-term  stability  of  the  frequency  spectrum  does 
not  exist.    This  observation  raises  some  doubt  as 
to  the  validity  of  spectral  measurements  which  have 
been  previously  reported  in  the  literature  for  use 
in  tissue  identification.    Most  researchers  using 
spectral  signatures  have  neither  considered  nor 
even  discussed  short-term  versus  long-term  stabili- 
ty problems.    In  using  the  A-scan  as  a  tissue  sig- 
nature, these  researchers  must  remember  that  all 
observations  in  vivo  are  of  a  dynamic  system.  Even 
with  held  inspiration,  the  organ  in  the  patient  is 
moving  throughout  the  cardiac  cycle  and,  in  typical 
clinical  ultrasound  laboratories  using  scanners  with 
hand-held  transducers,  further  movement  is  intro- 
duced through  the  inability  of  the  ultrasonographer 
to  hold  the  transducer  steady. 


~'j;^\^\/'^-~-'^~^\f\/^  sA^  ^^^.A^— -^-^N^^vVXy — ^  ^/V^ 


^-^Aa /V^--^\/Va^- — — — /-vysy^w'-'-  ^^\/^--^^>^^^v\A- 


Fig.  8,    Thirty-two  pulse-echoes  obtained  from  a 
carcinomic  region  of  the  same  kidney  en- 
sonified  in  preparing  the  data  shown  in 
figure  11.    Pulse-to-pulse  separation  is 
0.001  second. 


Fig.  9.    The  thirty-two  frequency  spectra  correspond- 
ing to  the  A-scans  shown  in  figure  11. 


306 


--W^/\/\a.^^yV5{/\/\A/V/%^  --v-^V—  -^^^ 

— \/\y\/\/^^v — v/\yV--N/\^ — ^yAyvx/^.-^^^ — - — ^- —  ■ 

^''^-^-^YA-yv — —^-7 — — ^XXAA/^-^^"^  ^~ — ^ — ~\^^\yx^v^\/v  

— — - — -^yY\7V'-V\/\AA^^-~-'^^wV\^~ — .^v — ^ 


Fig.  10.  The  thirty-two  frequency  spectra  corre-         Fig.  12.  Thirty-two  pulse-echoes  obtained  from 
spending  to  the  A-scans  shown  in  figure  12.  a  carcinomic  region  of  the  same  kidney, 

again  with  a  pulse-to-pulse  separation 
of  0.25  second. 


Fig.  n 


Thirty-two  pulse-echoes  obtained  on  a 
long-term  basis  from  a  normal  region  of 
the  same  kidney  ensonified  in  preparing 
the  data  shown  in  figure  11.  Pulse-to- 
pulse  separation  is  0.25  second. 


2.W1HI  3,9nHi 


Fig.  13.  The  thirty-two  frequency  spectra  corre- 
sponding to  the  A-scans  shown  in  figure  16. 


307 


Fig.  14.  The  thirty-two  frequency  spectra  corre- 
sponding to  the  A-scans  shown  in  figure  16. 


4.    Data  Reduction 

In  the  early  1970s  Mountford  and  Wells  [3,8,9] 
described  their  work  in  the  quantitative  analysis 
of  A-scans  taken  from  normal  and  cirrhotic  livers. 
Initially  their  data  was  obtained  from  oscilloscope 
photographs  of  the  rf  A-scan  obtained  from  the  sub- 
coastal  examination  of  supine  patients  through  the 
anterior  abdominal  wall  (the  same  physiological  ap- 
proach taken  in  our  own  work).    Thirty  A-scan  oscil- 
lograms were  recorded  per  patient  over  a  population 
of  30  normal  and  13  cirrhotics  using  an  ultrasonic 
frequency  of  1.5  MHz  (with  an  unrecorded  bandwidth). 
One  hundred  data  points  were  recorded  for  each  A- 
scan  over  a  time-span  of  65  microseconds  which  cor- 
responds to  approximately  4.8  centimeters  of  tissue 
(digitizations  started  at  approximately  5  centi- 
meters beyond  the  anterior  abdominal  wall).  Mea- 
sures were  made  which  related  to  signal  amplitude, 
the  rate  of  change  of  signal  amplitude,  frequencies, 
rise-times,  fall-times,  peak-to-peak  amplitudes, 
etc.    The  most  significant  of  these  measures  in 
separating  the  normal  from  the  cirrhotic  liver 
(using  Student's  t-test)  are  tabulated  below. 


Measure 

t-test 

Rate  of  mean  amplitude  decay 

1 

28 

Mean  echo  amplitude  at  intercept 

12 

38 

Mean  echo  amplitude  at  mid-point 

13 

31 

Mean  trough-to-peak  rise-time 

4 

20 

Mean  peak-to-trough  fall-time 

2 

39 

Mean  trough-to-peak  amplitude 

2 

07 

Total  number  of  troughs 

3 

55 

Total  number  of  peaks 

3 

59 

As  can  be  seen  the  most  successful  measure  is  the 
mean  echo  amplitude  at  the  mid-point  of  the  time- 
window.    The  associated  t-test  value  indicates  an 
extremely  high  success  rate  in  separating  normals 
from  cirrhotics  over  this  particular  patient  popula- 
tion.   Other  significant  measures  are  the  mean  rise- 
time  and  the  total  number  of  peaks  and  troughs.  It 
is  somewhat  surprising,  since  these  measures  direct- 
ly relate  to  the  frequency  of  the  signal,  that  the 
authors  state  that  "(the)  restricted  bandwidth  of 
the  system  used  in  the  present  study  appears  to 
prejudice  the  possible  use  of  frequency  spectrum 
analysis  as  a  diagnostic  index." 

Subsequent  to  the  above-reported  experiment  the 
authors  modified  their  experimental  approach,  dis- 
carded photographic  recording  of  the  A-scan,  and  de- 
veloped equipment  for  real-time  digitization.  In 
their  second  experiment  100  time-samples  were  extract- 
ed per  patient  the  sampling  interval  being  0.1  micro- 
second (corresponding  to  a  tissue  span  of  2  centi- 
meters).   The  initiation  point  of  their  digitizations 
was  6.5  centimeters  beyond  the  anterior  abdominal 
wall.    Ten  normals  and  three  cirrhotics  were  examin- 
ed with  180  A-scans  recorded  in  interrupted  quiet 
respiration  and  an  additional  180  A-scans  recorded 
at  held  maximal  inspiration  (except  for  cirrhotics  in 
which  recordings  were  made  only  at  held  maximal  in- 
spiration).   In  the  second  experiment  the  only  mea- 
sures reported  were  the  rate  of  mean  echo  amplitude 
decay  and  the  mean  echo  amplitude  and  intercept. 
The  frequency  utilized  was  1.67  MHz  and  it  can  be 
inferred  from  the  authors'  discussion  that  the  pulse 
duration  was  approximately  3  microseconds.    The  sam- 
pling aperture  was  12  ns  with  digitization  at  6-bits. 
Because  of  the  small  sample  size  the  t-test  was  not 
calculated,  but  it  was  found,  as  before,  that  the 
mean  echo  amplitude  at  mid-point  (for  held  maximal 
inspiration)  could  successfully  be  used  to  separate 
normals  from  cirrhotics.    Interestingly,  the  data 
spread  in  this  experiment  appears  to  be  far  greater 
than  in  the  initial  experiment  indicating  that  the 
earlier  low  error  rate  values  may  be  subject  to 
question.    After  these  first  two  experiments  were 
conducted,  no  further  investigations  apparently  have 
been  carried  out. 

Our  own  analysis  has  gone  beyond  that  of  Mount- 
ford  and  Wells  in  that  there  is  equal  emphasis  on 
both  frequency-domain  analysis  and  time-series 
analysis.    The  analysis  has  been  carried  out  accord- 
ing to  traditional  statistical  lines  (see,  for  ex- 
ample, Wilks  [10]).    This  approach  also  appears  to 
have  been  taken  by  Decker  et  al.  [11]  at  the  Uni- 
versity of  Bonn,  in  the  analysis  of  A-scans  taken 
from  the  eye.    In  our  own  work  we  consider  each  of 
the  five  regimes  of  our  data  as  a  tissue  signature 
and,  for  each  A-scan  recording  for  each  subject, 
make  the  same  statistical  measures.    The  signature 
is  represented  as  follows: 


Signature  =  f (x^ ) 


i  =  1,. . . ,128 


The  function  f  may  represent  any  one  of  the  five 
signal  regimes.    In  other  words,  it  may  be  either 
the  rf  recording,  the  rectified  rf,  the  envelope  of 
the  rf,  or  the  Wiener  spectrum  or  deconvolved  Wiener 
spectrum  of  the  rf. 

Measures  With  Respect  to  the  Argument 

Eight  measures  are  made  with  respect  to  the  argu- 
ment X.    Before  making  these  measures  the  function 
is  normalized  in  the  following  manner: 


308 


=  1 


This  normalization  is  applied  to  both  time-series 
and  frequency-domain  data  and  can  be  thought  of  as 
a  normalization  to  a  standard  acoustic  pressure  for 
the  time-series  data  and  to  a  standard  acoustic 
povver  for  the  frequency-domain  data.    After  normali- 
zation sunmations  are  carried  out  over  four  "bins" 
of  the  data  points  as  follov/s: 


N 


th 


■i  3M 


N  =  1 

n  =  2 

N  =  3 

N  =  4 


=  0-15 
=  16-31 
=  32-47 
=  48-63 


This  produces  the  first  4  measures  for  each  regime. 
Next  the  moments  of  f  are  calculated  as  follows: 


„th 


Moment  =  T.       -  0^f{x.)         N  =  2,3,4 


x.f(x.) 


The  above  equations  produce  8  measures  v/ith  re- 
spect to  the  argument  x.    Next  the  statistical 
properties  of  the  values  of  f  are  computed.  A 
histogram  or  probability  density  function  of  the 
values  of  f  is  calculated  and  called  p(yj)  v/here 
p  is  the  likelihood  that  f  has  the  value  yj.  The 
probability  density  function  is  normalized  to  0 
mean  and  unit  variance  as  follov/s: 


0 


1 


This  normalization  once  again  insures  that  arbitrary 
settings  in  acoustic  amplification  and  on  received 
power  levels  do  not  effect  the  results.    (Note  that, 
in  contradistinction  to  the  experiments  by  Mountford 
and  Wells,  we  do  not  operate  a  system  which  has 
open-loop  control  on  the  transmitted  and  received 
signal  levels  but,  rather,  operate  in  an  arbitrary 
level  mode).    Once  the  function  p  has  been  computed, 
we  record  its  third  and  fourth  moment  v/hich,  of 
course,  are  proportional  to  the  skew  and  kurtosis  of 
this  function,  respectively. 

Then  the  co-occurrence  matrix  or  diagram  is 
generated,  i.e.,  p(yj,y|^)  which  is  the  likelihood 
that  two  adjacent  values  of  f  have  precisely  the 
values  yj  and  y|^.    This  function  is  tv/o-dimensional 
and  may  be  represented  using  a  scatter  plot  of  the 
values  yj  and  y|^  as  is  shov;n  in  figure  15.    As  can 
be  seen  from  figure  15,  an  ellipse  may  be  fitted  to 
the  data  from  which  v/e  compute  eigenvectors  (the 
major  and  minor  axes),  eigenangle  (the  angle  made  to 
the  yj  axis),  and  the  eigenvolume  (area  of  the  el- 
lipse) . 

Finally,  the  3-gram  is  computed.    Since  there  are 
only  128  values  of  f,  a  three-dimensional  scatter- 
gram  is  sparsely  filled  from  a  statistical  point  of 
view  so  that,  rather  than  fitting  a  three-dimension- 
al ellipsoid,  an  8-bin  histogram  is  computed  with 
the  values  of  f  inarized  with  respect  to  a  threshold 
equal  to  the  mean  value  of  f  so  that  the  bins  of  the 
histogram  are  given  by 


(yj.yk^y^)  =  000,000,. ..,111 


Fig.  15.    The  di-gram  (scatter  plot)  corresponding 
to  a  single  A-scan  taken  from  a  normal 
liver  using  5-bit  digitization  over  28 
points  taken  at  100  ns  intervals. 


The  range  of  this  binary  3-gram  is  then  entered  as 
the  final  measurement.    This  leads  to  8  measures 
with  respect  to  the  argument  x,  2  measures  related 
to  first  order  probability,  3  measures  related  to 
second  order  probability,  and  1  measure  of  third 
order  probability.    Since  all  of  these  measures  are 
computed  for  each  of  the  five  data  regimes,  a  total 
of  14  measures  are  made  per  regime,  leading  to  70 
measurements  per  A-scan. 

5.    Biopsy  Tissue  Image  Analysis 

As  mentioned  in  Section  1,  biopsied  tissue  is  be- 
ing prepared  not  only  for  visual  examination  but  al- 
so computer  analysis.    Using  standard  4  m  thick 
tissue  slices  stained  with  hematoxyl in-eosin  and 
mounted  on  microscope  slides,  the  Automatic  Light 
Microscope  Scanner  (ALMS)  at  the  Jet  Propulsion 
Laboratory  produces  512  x  512  pixel  (picture  point) 
digitizations  of  the  previously  ensonified  sample 
as  shown  in  figures  16  and  17.    Each  image  field  is 
approximately  0.5  millimeter  square  selected  so  that 
a  statistically  sufficient  number  of  various  tissue 
cell  types  are  displayed.    A  software  system,  called 
AUTOPIC,  was  written  incorporating  cellular  logic 
commands  for  image  analysis  in  the  Cartesian  tes- 
sellation patterned  after  the  hexagonal  parallel 
pattern  transforms  of  Golay  [12].    These  transforms 
have  been  found  useful  by  Preston  [l3,14]  in  the 
analysis  of  images  of  human  white  blood  cells,  chest 
x-rays,  ophthalmol ogical  ultrasonograms,  etc.  Using 
AUTOPIC  applied  to  the  images  shown  in  figures  16 
and  17  normal  cell  nuclei  and  i nf 1 anmatory  cell 
nuclei  v;ere  located  (figs.  18  and  19)  using  a  com- 
mand sequence  which  caused  the  analysis  flow-charted 
in  figure  20  to  be  carried  out.    Since  an  increase 
of  inflammatory  cells  (called  "microcytes"  in  fig. 
20)  is  an  indicator  of  disease,  this  analysis  pro- 
vides an  index  of  histopathology  which  may  be  cor- 
related with  the  A-scan  measures  to  determine  the 
relevance  of  these  measures  in  assaying  disease 
entities. 


309 


Fig.  16.    Computer  generated  graytone  image 
of  a  0.5  X  0.5  mm  region  in  an 
eosin-hematoxyl in  stained  tissue 
section  from  a  normal  rabbit  kidney. 


Fig.  17.    Graytone  image  (0.5  x  0.5  mm)  of 

pyel onephri tic  rabbit  kidney  tissue. 


Fig.  18.    Computer  analysis  of  the  normal 
rabbit  kidney  tissue  image  shown 
in  figure  16  showing  an  overlay 
of  white  squares  and  white  circles 
to  indicate  nuclei  classified  as 
microcytes  and  macrocytes, 
respectively. 


Fig.  19.    Computer  analysis  of  the  pyelone- 
phritic  rabbit  kidney  tissue  image 
(fig.  17)  showing  a  microcytes  and 
macrocytes  as  in  figure  18. 


6.  Results 

Most  of  the  program  to  date  has  been  allocated  to 
both  the  construction  and  calibration  of  hardware 
and  to  the  development  of  a  software  system  for  both 
the  computation  and  graphical  display  of  the  data 
gathered.    In  this  paper  we  report  on  our  first 
human  biopsy  patient,  a  female  aged  64  years,  whose 
pathology  was  a  large  carcinoma  of  the  kidney.  By 
observation  of  the  B-scan  the  region  to  be  digitized 
is  readily  located.    Using  the  equipment  described 


in  Section  2,  the  A-scans  given  in  figures  7  through 
14  were  recorded. 

Thirty-two  of  these  A-scans  were  of  normal  kidney 
tissue  and  32  of  the  carcinoma.    Using  the  short- 
term  data  given  in  fig  ures  7  through  10,  all  70  mea- 
sures were  examined  for  the  64  recordings  and  the 
two  measures  most  useful  in  differentiating  normal 
and  pathological  tissue  were  found  to  be  the  second 
moment  with  respect  to  the  argument  x  where  x  was 
selected  in  the  frequency-domain  and  the  contents  of 
the  second  bin  (or  quartile)  of  the  rectified  time- 


310 


GRAY  SCALE 
IMAGE 


COPY 
[5]  ^ 


MASK 


SELECT  128 
WINDOW  [l] 


_  rBLOCKLj 

i]  — I — n  L 


THRESHOLD 
FOR  DARK 
OBJECTS  [2] 


REMOVE  SMALL 
DARK  OBJECTS 


FORM  MASK 
HFOR  LARGE  DARK 
OBJECTS  [5-6] 


THRESHOLD  FOR 
MODERATELY  DARK|_ 
OBJECTS  [9] 


REPLICATE  POSSIBLE 
M I CR0CY1ES 
[10-45] 


MICROCYTES 


REPLICATE  NON 
MACROCYTES 
[6  9-88] 


DECREMENT  AND 

CONTINUE  REPLI- 

COPY 

THRESHOLD 

CATION  OF  POSSIBLE 

RESULT 

AGAIN  [48] 

MICROCYTES  [49-64] 

[46] 

SUBTRACT 
[65-66] 


STEREO  TYPE 
MACROCYTE 
RESIDUES  [1  31  -  I  34] 


COUNT 

AND 
OUTPUT 
[136-1  37 


COPY  1 

SUBTRACT 

[I27]t— 

[129-130] 

REMOVE 
NOISE 
[68] 


REDUCE  TO 
RESIDUES 
[126-126] 


ERASE 
RESIDUES 
[128] 


SUBTRACT 

REMOVE 

REDUCE  IN 

[89-90] 

NOISE  [91  ] 

SIZE[i08] 

THRESHOLD  FOR 
MODERATELY 
DARK  OBJECTS  [92] 


OBTAIN 
MACROCYTES 
[12  4] 


FORM  MICROCYTE 
MASK  [121-123] 


COPY 

SUBTRACT 

COMBINE 

REDUCE  TO 
RESIDUES 

[10  3] 
f 

[1O5-IO6] 
t  - 

[109-  111] 

[112-114] 

REMOVE 
NOISE  [95-99] 


REDUCE  SMALL 
-H    OBJECTS  TO 

RESIDUEs[lOO-l02] 


ERASE 
RESIDUES 
[104] 


GENERATE 
MICROCYTE 
STEREO  TYPE 

[11  5-1  1  7] 


COUNT 

AND 
OUTPUT 
[119-120] 


Fig.  20.  Block  flow  chart  of  the  image  analysis  program  carried  out  in  the  digital 
computer  analysis  of  the  tissue  images  shown  in  figures  16  and  17. 


series.    Scatter  plots  of  these  measures  are  shown 
in  figure  21  for  the  32  A-scans  of  both  the  normal 
kidney  and  the  abnormal  kidney  under  short-term  con- 
ditions.   The  Mahalanobis  Distance  was  taken  as  a 
measure  of  separability  and  indicates  a  high  degree 
of  confidence  in  telling  these  two  types  of  tissues 
apart.    However,  when  the  A-scan  data  taken  under 
long-term  conditions  (figs.  11  through  14)  were 
utilized,  the  Mahalanobis  Distance  indicated  some- 
what worse  differentiation  between  the  two  types 
of  tissue  (fig.  22)  typifying  the  effects  of  the 
dynamic  variability  of  the  A-scan. 

The  other  experiment  which  is  reported  here  in- 
volved the  introduction  of  pyelonephritis  in  the 
animal  kidney  and  the  comparative  analysis  of  A- 
scan  data  and  histologic  data  derived  from  images 
of  biopsies  of  the  ensonified  tissue.    The  animal 
model  selected  was  the  rabbit. 

The  disease  was  introduced  unilaterally.    At  the 
acute  stage,  four  weeks  after  introduction  of  the 
disease,  the  kidneys  (both  normal  and  abnormal) 
were  operatively  exposed  and  128  A-scans  were  re- 
corded for  each  organ  via  a  6  cm  water  bath  coupler 
suggested  by  Dr.  Goans  of  the  Oak  Ridge  National 
Laboratory.    This  avoided  signal  corruption  by  in-  - 
tervening  tissue. 

Results  of  the  A-scan  analysis  (for  animal  R06) 
are  shown  in  figure  23,  which  is  a  scattergram  of 
the  data  on  256  A-scans  taken  in  groups  of  32  in  8 
second  intervals  over  several  minutes  using  the 
two  best  computer  measures  for  separating  normal 
from  abnormal.    The  Mahalanobis  Distance  is  3.2 
and  the  classification  success  rate  is  98  percent. 
Similar  results  (100%  classification)  were  ob- 
tained for  the  tissue  image  analysis  (fig.  24)  using 
a  single  measure,  i.e.,  the  count  of  microcyte  den- 
sity in  the  four  0.25  x  0.25  mm  regions  shown  in 
figures  17  and  19.    Pictorial  and  A-scan  data  taken 
from  six  other  animals  is  now  being  analyzed. 


'  .0  75.0        150.0      225.0      300.0      375.0       450. 0  525.0 

PSM2 

104:  Tumor  vs.  Normal  (32  ms  ) 

Fig.  21.  Scatter  plot  of  64  points  in  a  2-coordinate 
space  representing  the  best  measures  for 
differentiating  ultrasonic  pulse  echoes 
from  those  received  from  carcinomic  human 
kidney  tissue  in  vivo.    The  coordinates 
are  the  variance  of  the  power  spectrum 
(PSM2)  and  the  second  quartile  sum  of  the 
rectified  waveform  (RED2).    The  plot  is 
derived  from  the  short-term  data  shown  in 
figures  7-10. 


311 


Fig. 


75.0         150.0       225.0       300.0       375.0       450.0  525.0 

PSMZ 

104:  Tumor  us.  Normal 

22.  Scatter  plot  similar  to  figure  23  using 
data  extracted  from  figures  11-14  showing 
A-scan  information  gathered  in  the  long- 
term  (8  seconds).    Dynamic  variance  of  the 
data  causes  the  classification  success  to 
drop  to  87  percent  for  normal  human  kidney 
tissue  and  to  91  percent  for  carcinomic 
human  kidney  tissue. 


S- 


J  I  I     I  I 


^  Normal 
I   I  Abnormal 


55 


220 
Number  of 
microcytes 
per  quadrant 


Fig.  24. 


Histogram  based  on  the  microcyte  count  per 
quadrant  (0.25  x  25  mm)  in  figures  18  and 
19  showing  100  percent  success  in  computer- 
ized differentiation  between  normal  and 
pyelonephri tic  rabbit  kidney  tissue  images. 
The  microcyte  count  is  indicative  of  in- 
flammation and  is  significantly  higher  in 
pyel onephri  ti  s . 


□  o 


log 


log(Pr,D4) 


n06:  Pyalo  vs.  Normal 


Fig.  23.  Scatter  plot  of  256  points  using  the  two 
measures  which  were  found  to  be  best  for 
distinguishing  ultrasonic  echoes  from 
normal  and  pyelonephritic  rabbit  kidney 
in  vivo.    These  measures  are  the  fourth 
moment  of  the  power  spectrum  (PSD4)  and 
the  variance  (second  moment)  of  the  rec- 
tified signal  data  histogram.  Classifi- 
cation success  for  this  long-term  data 
is  100  percent  for  normal  tissue;  93 
percent  for  pyelonephritic. 

7.    Acknowledgements  _ 

In  addition  to  the  workers  and  colleagues 
acknowledged  in  the  text,  the  authors  would  like 
to  acknowledge  the  work  of  Dr.  Niel  Wald,  Chairman 
of  the  Department  of  Radiation  Health,  University 
of  Pittsburgh  and  the  staff  of  his  Biomedical  Image 
Processing  Group,  Jet  Propulsion  Laboratories  for 
assistance  in  image  analysis  and  recording;  Dr. 
Donald  Eitzen,  Ultrasonic  Standard  Group,  National 
Engineering  Laboratory,  National  Bureau  of  Standards, 
and  his  staff  for  performing  the  calibration  of  our 
tranducer;  Dr.  Frank  Fry,  Ultrasound  Research 
Laboratories,  Indianapolis  Center  for  Advanced 
Research,  and  his  students  for  calibration  of  the 
beam  pattern  of  the  transducer  and  measurements  of 
its  performance  using  phantoms;  Drs.  Andrew  Dekker, 
Denis  Borochovitz,  Jeffrey  D.  Hubbard,  University 
of  Pittsburgh  Health  Center,  for  visual  reading 
and  reporting  on  tissue  section  pathology;  Dr. 
Terrance  Matzuk,  consultant,  for  providing  ultra- 
sound interface  electronics;  Mr.  Gilbert  Arnold, 
Mellon  Institute,  and  his  staff  for  providing  the 
illustrations;  Mrs.  Tanya  Rogers,  Department  of 
E-lectrical  Engineering,  Carnegie-Mellon  University 
for  typing  the  manuscript. 


312 


References 

[1]    Wild,  J.  J.  and  Reid,  J.  M.,    Application  of 
echo  ranging  techniques  to  the  determination 
of  structure  of  biological  tissue.  Science 
115.  226  (1952). 

[2]  Fry,  E.,  Okuyama ,  D. ,  and  Fry,  F.  J.,  Ultra- 
sonic differentiation  of  normal  liver  struc- 
ture as  a  function  of  age  and  species,  Proc. 
6th  Intern' 1.  Cong,  on  Acoustics,  Tokyo  (1968). 

[3]    Mountford,  R.  A.  et  al . ,  Ultrasonic  liver 

scanning:    automatic  A-scan  analysis.  Physics 
in  Medicine  and  Biology  17,  559-569  (1973). 

[4]    Ossoinig,  K.  C,  Quantitative  echography  - 
the  basis  of  tissue  differentiation,  J.  Clin. 
Ultrasound  1,  190  (1973). 

[5]    Fields,  S.  and  Dunn,  F.,  Correlation  of  echo- 
graphic  visualizability  of  tissue  with  bio- 
logical composition  and  physiological  state, 
J.  Acoust.  Soc.  Amer.  54,  809  (1973). 

[6]    Taylor,  K.  J.  W.  and  Milan,  J.,  Digital  A- 
Scan  Analysis  in  the  Diagnosis  of  Chronic 
Splenic  Etilargement,  in  Ultrasonic  Tissue 
Characterization,  M.  Linzer,  ed..  National 
Bureau  of  Standards  Special  Publication 
453,  pp.  71-80  (U.S.  Government  Printing 
Office,  Washington,  D.C.,  1976). 

[7]    Preston,  K. ,  Jr.,  Use  of  Pattern  Recognition 
for  Signal  Processing  in  Ultrasonic  Histo- 
pathology,  in  Ultrasonic  Tissue  Characteriza- 
tion, M.  Linzer,  ed..  National  Bureau  of 
Standards  Special  Publication  453,  pp.  51-60 
(U.S.  Government  Printing  Office,  Washington, 
D.C.,  1976). 


[8]    Mountford,  R.  A.  and  Wells,  P.  N.  T.,  Ultra- 
sonic liver  scanning:    the  quantitative 
analysis  of  the  normal  A-scan,  Physics  in 
Medicine  and  Biology  17,  14-24  (1972a). 

[9]    Mountford,  R.  A.  and  Wells,  P.  N.  T.,  Ultra- 
sonic liver  scanning:    the  A-scan  in  cir- 
rhosis. Physics  in  Medicine  and  Biology  17, 
261-269  (1972b). 

[10]    Wilks,  W.,  Mathematical  Statistics,  Wiley 
(1963). 

[11]    Decker,  D. ,  Epple,  E.,  Leiss,  W.,  and  Nagel , 
M.  ,  Digital  computer  analysis  of  time- 
amplitude  ultrasonograms  from  the  human  eye, 
J.  Clin.  Ultrasound  1  (2),  150  (1973). 

[12]    Golay,  M.  J.  E. ,  Hexagonal  parallel  pattern 
transforms,  IEEE  Trans.  Comput.  C-18,  733 
(1969) . 

[13]    Preston,  K. ,  Jr.  and  Onoe,  M.  ,  Digital  Proc- 
essing of  Biomedical  Images,  Plenum  Press 
(1976).  ■ 

[14]    Preston,  K.,  Jr.,  Application  of  the  Golay 
Transform  to  Image  Analysis,  in  Digital 
Image  Processing  and  Analysis,  1' Institute 
de  Recherche  d  Informatique  et  d ' Automatique , 
Paris  (1976). 


313 


CHAPTER  10 
TISSUE  VIABILITY  AND  TISSUE  PHANTOMS 


315 


i 


! 
i, 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


DAMAGE  AND  DEATH  IN  TISSUES  AND  ASSOCIATED  CHANGES  IN  THEIR  MECHANICAL  PROPERTIES 


L.  Weiss 

Roswell  Park  Memorial  Institute 
Buffalo,  New  York    14263,  U.S.A. 


Mechanical  effects  of  tissue  damage  associated  with  artefactual  change  during 
examination,  and  that  developing  naturally  during  tumor  development,  have  been  de- 
monstrated by  a  quantitated  cell  detachment  test.    Although  the  precise  relationship 
between  this  semi-destructive  test  of  tissue  properties  and  ultrasonic,  nondestruc- 
tive tests  is  uncertain,  it  appears  feasible  to  consider  the  following  points  In 
studies  of  the  interaction  of  ultrasound  with  tissue  specimens:    a)  thin  specimens 
should  be  examined  quickly  to  avoid  anoxic  and  hypoxic  damage;  b)  note  must  be  made 
of  the  heterogeneities  within  solid  tumors,  particularly  those  produced  by  necrosis; 
c)  degenerative  changes  in  a  tumor  and  surrounding  non-malignant  tissues  may  act  as 
image-enhancers . 

Key  words:    Cell  detachment;  mechanical  properties;  tissue  damage;  tissue  death. 


1.    Introduction  2.    Cell  Detachment  from  Tissues 

The  interactions  of  ultrasound  with  tissues.  Cylinders  are  punched  from  tissues  using  a  13 

depends  in  a  sense  on  their  mechanical  properties.     gauge  ("^  2  mm)  trocar  and  cannula,  and  these  are 

then  divided  into  1  or  2  mm  lengths  over  a  scale, 
under  magnification.    Two  such  cylinders  are  placed 
into  each  of  a  number  of  screw-capped,  glass  vials, 
4.5  cm  long  and  1.3  cm  diameter,  containing  2  ml 
of  cold  Hanks'  balanced  saline  solution.  These 
operations,  which  take  20  to  30  minutes,  are  all 
done  in  solutions  at  4  °C  to  minimize  metabolic 
di  fferences . 

The  vials  are  clamped  onto  a  reciprocating 
shaker,  making  275  oscillations  per  minute,  with 
an  excursion  of  4.5  cm,  for  40  minutes,  at  room 
temperature  (25  °C).    In  some  experiments,  tissue 
cylinders  were  shaken  for  10  minutes  only.  After 
shaking,  0.25  ml  of  10  percent  buffered  formalde- 
hyde is  added  to  each  vial  and  mixed  by  gentle 
inversion.    When  present,  the  macroscopic  remains 
of  the  cylinders  rapidly  settle  to  the  bottoms  of 
the  vials,  and  the  released  cells  and  cell  clusters 
in  the  supernatant  fluid  are  counted  in  a  Fuchs- 
Rosenthal  chamber,  and  their  diameters  measured 
with  an  image-splitting  eyepiece  (Vickers-A. E. I . , 
England)  at  a  final  magnification  of  X300.    It  has 
previously  been  demonstrated  that  the  described 
formaldehyde-fixation  does  not  affect  these  two 
numerical  determinations. 

An  estimate  of  the  volume  of  tissue  liberated 
by  shaking  was  obtained  from  the  product  of  the 
cube  of  the  mean  "particle"  radius  and  the  mean 
number  of  such  cell  units.    Volume  functions  of 
this  type  are  used  for  purposes  of  comparison. 

Small  pieces  of  tissue  were  fixed  in  formalin 
and  subsequently  examined  in  4  ym  stained  sections. 
Aliquots  of  liberated  cells  are  centrifuged  onto 


In  this  presentation,  tv;o  facets  of  tissue  damage 
will  be  examined  from  a  mechanical  viewpoint. 
The  first  is  a  consideration  of  the  hypoxia  which 
results  and  its  possible  artefactual  effects,  when 
pieces  of  tissue  are  examined.    The  second  is  a 
consideration  of  the  mechanical  heterogeneity  in 
and  around  solid  cancers. 

The  mechanical  properties  of  tissues  depend 
partially  on  the  properties  of  their  individual 
cells  and  partially  on  the  properties  of  the 
material  lying  between  them,  in  much  the  same  way 
as  the  properties  of  a  brick  wall  depends  on  the 
properties  of  the  bricks,  the  mortar  and  the  bonds 
between  them.    Tissue  properties  may  be  measured 
by  nondestructive  techniques,  such  as  those  in- 
volving ultrasound,  or  by  destructive  techniques 
which  result  in  reduction  of  the  tissue  into  either 
subcellular  dimensions,  or  into  its  component  liv- 
ing cells.    The  release  of  living  cells  from  tis- 
sues have  been  studied,  particularly  solid  cancers, 
as  part  of  a  program  on  the  mechanisms  of  metas- 
tasis [1-3]^.    In  this  work,  quite  sensitive 
quantitative  techniques  have  been  developed  to 
measure  cell  detachment,  which  is  dependent  on  the 
strength  of  the  intercellular  material.    In  tissues, 
the  intercellular  material  represents  a  modifica- 
tion of  the  extramembranous ,  peripheral  regions  of 
cells  which  is  rich  in  glycoproteins. 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


317 


slides  (Cytospin:    Shandon-Ell iott,  England),  and 
morphologically  characterized  after  reaction  with 
Wright's  stain.    A  full  description  of  the  tech- 
nique and  its  use  is  given  elsewhere  [4]. 

The  cell  detachment  test  described  here  is  non- 
destructive as  far  as  the  component  cells  of  a  tis- 
sue are  concerned,  and  their  viscoelastic  properties 
must  play  a  role  in  cell  detachment.    However,  as- 
suming that  the  plane  of  detachment  lies  within  the 
cell  periphery/intercellular  region,  then  the  tests 
are  destructive  for  this  tissue  component  in  the 
sense  that  they  are  essentially  concerned  with  ob- 
taining a  "yield-point".    In  contrast,  the  ultra- 
sonic techniques  discussed  here  are  completely  non- 
destructive.   Thus,  although  the  precise  relation- 
ship of  the  two  techniques  is  problematical  and  the 
results  are  expected  to  be  somewhat  different  in 
general  terms  it  would  be  expected  that  changes  in 
tissues  detected  by  one  procedure  might  well  be  re- 
flected in  the  other.    Proposed  work  will  hopefully 
clarify  this  relationship,  which  at  present  is 
simply  presented  as  a  working  hypothesis. 

3.    Examination  Artefacts 

In  determining  parameters  of  the  interactions 
of  ultrasound  with  tissues,  the  tissues  them- 
selves are  held  in  a  tank  of  fluid  in  the  ultra- 
sonic beam,  for  varying  periods  of  time.  From 
many  biologically  oriented  studies,  it  is  well- 
known  that  unless  the  fluid  is  of  physiologic 
tonicity  and  pH,  and  that  unless  the  tissues  them- 
selves are  well -oxygenated ,  tissue  damage  will 
occur.    The  relevant  question  is,  how  much  arte- 
factual  alteration  will  be  produced  in  the  ultra- 
sound/tissue interaction  by  this  damage? 

From  a  practical  viewpoint,  it  is  unnecessary 
to  investigate  the  effects  of  unphysiologic  fluids 
on  the  interactions,  because  many  solutions  are 
available  which  maintain  optimal  conditions  of 
ionic  strength  and  constitution,  pH  etc.  [5]. 
However,  one  outstanding  problem  is  the  possibility 
of  inadequate  tissue  oxygenation,  and  the  con- 
sequences of  it.    In  the  case  of  isolated  tissue 
slices,  Warburg  [6]  showed  that  their  limiting 
thickness  in  cm  (H)  for  O2  consumption,  where  D 
is  the  diffusion  coefficient,  A  is  the  rate  of 
respiration  of  the  tissue  (ml  O2  uptake/ml 
tissue/minute),  and  [O2]  is  the  oxygen  concentra- 
tion in  the  atmosphere  surrounding  the  tissue, 
then: 

H  =  (8-D-[02]/A)''=  . 

Thus,  in  pure  oxygen,  the  maximum  thickness  of  tis- 
sue permitting  the  passive  penetration  of  O2  to  its 
center  is  0.47  mm.    In  air  ([O2]  =  21  percent)  the 
corresponding  maximal  thickness  of  tissue  is  only 
0.21  mm.    Similar  considerations  have  to  be  taken 
into  account  in  determining  the  depth  of  fluid 
covering  examined  tissues.    Studies  on  the  kinetics 
of  gas  diffusion  in  cell  culture  systems  by  Mc- 
Limans  et  al .  [7]  have  demonstrated  that  as  pre- 
dicted by  Fick's  Law,  the  passive  diffusion  of 
oxygen  through  unstirred  fluid  to  actively  metabo- 
lizing tissues  is  very  sensitive  to  the  depth  of 
fluid,  and  that  in  the  case  of  isolated  liver  cells 
covered  by  only  1.6  mm  of  physiologic  saline,  the 
total  initial  equilibrated  concentration  of  oxygen 
will  be  exhausted  in  only  1280  seconds,  normal  bio- 
logic syntheses  will  be  impaired  and,  if  left  under 
these  conditions,  the  cells  will  die. 


In  order  to  get  a  feeling  for  ti ssue-oxygenation 
under  the  conditions  in  which  ultrasonic  tissue 
characterization  experiments  will  be  made,  freshly 
isolated  cylinders  of  rat  liver  1  cm  in  diameter 
and  0.5  cm  in  length,  were  obtained  from  exsan- 
guinated rats.    These  were  then  placed  in  a  beaker 
containing  100  ml  Hanks'  balanced  saline  solution, 
a  commonly  used  mammalian  "physiologic"  solution, 
buffered  at  pH  7.2,  and  maintained  at  37  °C.  The 
PO2  was  measured  at  various  depths  in  the  liver 
cylinder  determined  with  a  micrometer,  with  oxygen 
ultramicroelectrodes  having  a  tip  diameter  of 
15  pm.    The  electrode  has  been  described  elsewhere 
[8].    It  was  shown  that  in  the  five  minutes  re- 
quired to  start  measurements  after  removal  of  the 
liver  cylinders,  the  oxygen  tensions  at  depths  of 
1  to  2.5  mm  were  near  zero,  and  at  a  depth  of  0.5 
mm,  they  were  approximately  60  mmHg.  These 
measurements  compare  with  mean  values  for  the  liver 
in  the  unanesthetized,  spontaneously  respiring  rat 
of  20  to  30  mmHq  [9].    Thus,  there  is  good  reason 
for  considering  the  effects  of  hypoxia  in  ultra- 
sonic experiments,  since  in  most  of  these,  com- 
paratively thick  pieces  of  tissues  are  used. 

It  has  been  shown  that  lysosomal  changes  in 
rat  hepatic  parenchymal  cells  are  induced  by  sub- 
jecting animals  to  hypoxia  [10].    It  has  also 
been  shown  that  activation  of  lysosomes  with  re- 
lease of  lysosomal  enzymes  into  the  tissues  can 
cause  massive  tissue  degradation  [11]  within 
hours,  and  changes  detectable  by  cell  detachment 
tests  on  cellular  monolayers,  within  2  minutes 
[12]. 

In  order  to  determine  if  the  treatment  of  the 
1  cm  diameter  cylinders  leading  to  anoxia  or 
hypoxia  resulted  in  analogous  changes,  shaking- 
tests  of  the  type  described  earlier  were  made  on 
this  material.    After  incubation  in  Hanks'  solu- 
tion at  37  °C  for  periods  up  to  2  hours,  plugs 
were  cut  from  the  1  cm  cylinders,  and  the  volumes 
of  cells  released  at  the  different  times,  from  the 
outermost  1  mm  regions  near  to  the  cut  surface  and 
the  innermost  2  mm  thick  regions  at  the  center  of 
the  5  mm  thick  cylinder  were  compared.    The  re- 
sults shown  in  table  1  indicate  that  greater 
(.02  >  p  >  .01)  volumes  of  parenchymal  cells  are 
released  from  the  inner  anoxic  region  at  the 
center  of  liver  cylinders  than  from  the  outer 
parts.    During  the  experiments,  detectable  amounts 
of  protein  and  acid  phosphatase  (a  lysosomal  hydro- 
lase) are  released  into  the  medium,  but  this  facet 
of  the  experiments  needs  developing. 

It  should  be  noted  that  examination  of  standard 
histologic  preparations  of  the  liver  revealed  no 
clear-cut  evidence  of  damage  after  2  hours  incuba- 
tion, and  would  not  have  permitted  predictions  of 
the  detachment  experiments.    It  is  of  interest 
that  Hueter,  in  experiments  cited  by  Dunn  [13]  was 
unable  to  detect  changes  in  the  ultrasonic  absorp- 
tion coefficient  of  liver,  measured  over  the  range 
of  1  to  6  MHz,  until  9  hours  after  death,  when  it 
progressively  decreased.    It  is  difficult  to  com- 
pare these  experiments  with  those  reported  here, 
since  Hueter 's  were  made  at  25  °C  (c.f.  37°)  and 
the  tissue  was  maintained  at  10  °C  between  measure- 
ments, and  it  is  well-known  that  reduction  in  tem- 
perature retards  post-mortem  change.    Future  work 
will  hopefully  clarify  the  correlation  between 
changes  detected  by  the  shaking  test  and  ultra- 
sound. 

If  degenerative,  autolytic  changes  due  to 
lysosomal  mobilization  prove  to  be  an  inconvenience 


318 


Table  1.    Release  of  liver  parenchymal  cells. 

Incubation      Sites       Numbers  of  cells        Unit  diameters  Comparative 

time                        released  (xlOOO)            (pm)  +  SE  volumes  released 

(minutes)                          ±  SE  (n)                      (n)  Inner  :  Outer 


Experiment  1 


30 

Inner 

57 

+ 

7. 

3 

(13) 

10.5 

+ 

0. 

14 

(161) 

1 

1 

8 

30 

Outer 

37 

+ 

4. 

9 

(16) 

10.0 

+ 

0. 

13 

(110) 

60 

Inner 

20 

+ 

3 

5 

(14) 

10.0 

+ 

0. 

15 

(131) 

1 

1 

4 

60 

Outer 

15 

+ 

1 

5 

(13) 

9.9 

+ 

0 

13 

(107) 

120 

Inner 

16 

+ 

0 

7 

(8) 

10.5 

+ 

0. 

39 

(100) 

1 

1 

5 

120 

Outer 

12 

+ 

1 

9 

(12) 

10.0 

+ 

0 

27 

(105) 

Experiment  2 

30 

Inner 

16 

+ 

2 

8 

(8) 

10.0 

+ 

0 

11 

(102) 

1 

1 

9 

30 

Outer 

10 

+ 

2 

7 

(9) 

9.5 

+ 

0 

11 

(94) 

60 

Inner 

15 

+ 

4 

0 

(6) 

9.6 

+ 

0 

13 

(136) 

1 

1 

2 

60 

Outer 

13 

+ 

3 

4 

(6) 

9.4 

+ 

0 

08 

(154) 

120 

Inner 

13 

+ 

1 

9 

(9) 

9.9 

+ 

0 

09 

(111) 

1 

1 

.5 

120 

Outer 

10 

+ 

2 

4 

(8) 

9.4 

+ 

0 

12 

(98) 

in  ultrasonic  measurements,  it  is  worthy  of  note 
that  the  analogous  changes  observed  in  vitro  [11, 
12]  were  inhibited  and/or  retarded  by  adding  small 
quantities  (c.  10  yg/ml )  of  hydrocortisone  hexi- 
succinate  to  the  medium. 

4.    Mechanical  Heterogeneity  in  Tumors 

The  essence  of  the  successful  treatments  of 
cancer  is  early  diagnosis.    In  the  breast  for  ex- 
ample it  can  be  seen  (fig.  1)  that  depending  on 
the  clinical  type  [14],  the  probability  of  metas- 
tasis varies  considerably  with  the  size  of  the 
cancer  at  the  time  of  diagnosis.    In  the  most 
malignant  (Type  A)  the  slowest  rate  of  increased 
probability  of  metastasis,  where  therapeutic  in- 
tervention would  be  most  effective,  is  in  lesions 
of  1  to  7  mm  diameter.    In  the  least  malignant 
(Type  B),  there  is  an  increased  probability  of 
metastasis  from  22  to  36  percent,  as  the  lesion 
increases  in  diameter  from  1  to  5  mm.    It  is  there- 
fore pertinent  to  ask  whether  changes  in  or  around 
a  cancer  as  it  develops  can  so  modify  the  mechani- 
cal properties  of  the  cancer  itself  or  the  sur- 
rounding nonmalignant  tissues,  that  the  image  of 
the  tumor  is  enhanced  and/or  changes  in  the  sur- 
rounding tissues  effectively  increase  the  "lesion" 
size. 

From  previous  considerations  of  examination  of 
artefacts,  the  question  arises  of  whether  various 
tissue  interactions  caused  by,  or  resulting  in  de- 
generative change  in  a  solid  cancer  could  fill  this 
role  of  "image  enhancer".    Cell  death,  or  necrosis 
is  a  salient  feature  of  many  solid  cancers.  One 
contributory  cause  of  this  is  that  blood  vessels 
entering  or  leaving  cancers  from  their  peripheries, 
tend  to  be  occluded  by  tissue  pressures  associated 
with  growth,  leading  to  vascular  stasis  and  throm- 
bosis.   The  inward  diffusion  of  nutrients,  and  the 
outward  diffusion  of  metabolic  products  becomes 
inadequate,  and  tissue  death  results.    There  are 


other  contributory  causes  of  necrosis  in  and  around 
tumors;  including  cell  and  humorally  mediated 
cytotoxicity  associated  with  inflammatory  change, 
but  the  end  result  is  the  same.    In  a  cancer  with 


5 

4- 


3  - 


1 

0.8 
0.6 


0.2 


O.lL 


type  B  , 


type  A 


Fig.  1. 


0         20        40         60         80  100 
Probability  of  distant  metastasis 

The  probability  of  distant  metastases  in 
breast  cancers  of  types  A  and  B,  of  dif- 
ferent size  at  the  time  of  diagnosis. 


319 


well-developed  necrosis,  the  necrotic  material 
liquefies  and  the  lesion  becomes  cystic.    The  dif- 
ferent absorptive  properties  of  the  liquefied  and 
solid  regions  lead  to  image  enhancement--as  is 
well-recognized  on  a  macroscopic  scale,  and  rep- 
resents an  example  of  the  heterogeneity  of  tumors 
[15]. 

Much  previous  work  has  not  taken  into  account 
the  heterogeneity  of  normal  ,  and  particularly 
pathologic  tissues.    This  heterogeneity  may  be  due 
to  normal  anatomic  structures  such  as  large  blood 
vessels,  fascial  planes  etc. ,  or  to  local  patho- 
logical situations  such  as  necrosis,  cyst  forma- 
tion, cellular  infiltrations,  tissue  edema  etc. 
Very  few  pathologic  lesions  consist  of  a  "pure" 
tissue  type,  and  for  this  reason  average,  integrat- 
ed interaction  data  collected  from  a  large  volume 
of  tissue  may  be  inappropriate  for  tissue  signa- 
ture studies.    In  any  general  consideration  of 
ultrasonic  tissue  characterization  it  is  mandatory 
to  consider  not  only  the  macroscopic  or  average 
features  of  lesions,  but  also  their  individual 
"microscopic"  components.    By  "microscopic",  a 
region  of  a  lesion  is  implied  which  has  the  dimen- 
sions of  the  narrowest  ultrasonic  beam  used  for 
tissue  characterization  experiments.    With  this  in 
mind,  the  mechanical  properties  of  different 
regions  of  the  same  tumor,  and  the  nonmalignant 
tissues  around  it  have  been  determined  by  the  cell 
detachment  technique,  on  Walker  256  "carcinosar- 
comas" transplanted  in  livers  of  rats. 

A.  Effect  of  Site  Within  Tumor 

Cystic,  subcutaneous  Walker  256  tumors  were 
selected,  with  necrotic  centers  of  approximately 
3  cm  mean  diameter,  and  with  wall  thicknesses  of 
2  to  5  mm  of  "healthy"  cancerous  tissue.    The  re- 
lease of  cells  from  the  inner  and  outer  1  mm 
lengths  of  cylinders  taken  radially  through  the 
cyst  walls  are  shown  in    table  2.    In  the  two  rep- 
resentative examples  shown,  3  and  6  times  greater 
volumes  of  tumor  were  released  from  the  inner, 
juxta-necrotic  regions  of  the  cyst  walls  than  from 
their  outer  parts. 

B.  Effects  of  Necrotic  Extract 
on  Tumor  Cel 1  Release 

The  subcutaneous  tumors  from  which  material 
was  obtained  fell  into  two  groups.    The  first  were 
roughly  spherical,  with  diameters  of  approximately 
1  cm,  and  contained  minimal,  friable  necrotic 
cores.    The  second  group,  were  roughly  egg-shaped 
with  diameters  of  approximately  4  cm  and  6  cm  in 


Table  2. 

Tumor  cel 1 s 

released  from  inner  and  outer 

parts  of  walls  of  subcutaneous 

cystic 

Walker  256 

tumors . 

"Cel 1 "- 

1  c  I  era  J  c 

1  itrdil   b  I  Zc 

Vol ume 

(X  1000) 

±  SE  (n) 

ym  ±  SE  (n) 

ra  1 1 0 

Inner 

Outer 

Inner  Outer 

Inner/Outer 

1  mm 

1  mm 

1  mm          1  mm 

202  +  65 

77  +  12 

7"7j.O     CT        1 A     ^     ")  1 
11     ±J.D  /4±J.I 

3:1 

(9) 

(9) 

(99)  (99) 

90  +  26 

36  ±  8.2 

110  +  3.3    83  ±  3.5 

6:1 

(9) 

(10) 

(100)  (100) 

the  short  and  long  axes  respectively,  and  were 
largely  necrotic,  with  healthy  cancerous  "rims" 
varying  from  2  to  4  mm  in  depth. 

Cylinders  obtained  from  the  peripheral  regions 
of  the  tumors  were  first  incubated  with  1/10  dilu- 
tions of  necrotic  extracts  in  HBSS  for  20  minutes 
at  37  °C  and  then  shaken  for  10  minutes  at  room 
temperature  (24  °C).    The  representative  results 
of  6  separate  experiments  with  3  separate  batches 
of  necrotic  extract  are  summarized  in  table  3. 
In  the  case  of  the  1  cm  diameter  tumors,  8  to  11 
times  greater  volume  of  tumor  was  released  after 
exposure  to  necrotic  material  than  in  the  appro- 
priate controls.    In  the  case  of  the  largely 
necrotic  tumors,  it  is  seen  that  pretreatment 
with  necrotic  extract  produced  no  significant 
increase  in  the  volume  of  tumor  released  by  shak- 
ing, compared  with  controls. 

C.    Effects  of  Tumor  on  Surrounding 
"Normal"  Tissues 

Following  the  direct  injection  of  Walker 
ascites  cells  (10^  cells  in  0.2  ml  HBSS)  into  the 
liver,  tumors  of  approximately  1  cm  diameter  grew 
within  10  days.    The  tumors  and  the  surrounding 
liver  were  bisected,  and  cylinders  of  liver  were 
removed  at  its  junction  with  the  tumor,  and  at 
0.5  and  1.0  cm  from  this  interface.    The  results 
of  10  separate  experiments,  together  with  re- 
lease data  from  2  normal  livers  are  given  in  table 
4.    It  is  shown  that  a  greater  volume  of  paren- 
chymal cells  are  released  from  tumor-bearing  than 
normal  livers.    In  addition  the  closer  the  normal 
liver  samples  are  to  the  tumor  interface,  the  more 
readily  are  cells  detached  from  them. 


Table  3.    Tumor  cylinders  incubated  with  necrotic  extract 

for  20  minutes  at  37  °C;  then  shaken  for  10  minutes. 


Exper imenta' 

(E) 

Control 

(0 

Tumors 

Numbers 

CI  ump 

Numbers 

Clump 

Volume 

rel eased 

diameter 

rel eased 

diameter 

ratio 

(X  1000) 

(ym) 

(X  1000) 

(ym) 

E:C 

±  SE  (n) 

+  SE  (n) 

±  SE  (n) 

±  SE  (n) 

c.  1  cm 

171  ±  56 

57  ±  2.0 

245  ±  16 

23  ±  0.3 

10:1 

(17) 

(100) 

(17) 

(100) 

>  4  cm 

376  +  56 

23  ±  1.0 

477  ±  61 

21  ±  0.6 

1.0:1 

(10) 

(199) 

(10) 

(197) 

320 


Table  4.    Release  of  "normal"  liver  parenchymal 
cells  surrounding  tumors  as  function 
of  distance  from  tumor  interface. 


Distance 
from 
tumor  edge 

Numbers 
released 
(X  1000) 
±  SE  (n) 

CI  ump 
size 
(ym) 
±  SE  (n) 

Vol ume 
ratio 
(cf.  normal) 

0  cm 

324  ±  16 
(65) 

42  ±  0.6 
(510) 

4.1:1 

0.5  cm 

256  ±  20 
(36) 

43  ±  0.6 

(306) 

3.7:1 

1.0  cm 

224  ±  17 
(19) 

38  ±  0.5 
(204) 

2.3:1 

Normal 
liver 

350  ±  16 
(84) 

25  ±  0.2 
(400) 

1.0 

D. 


Effects  of  Necrotic  Extract  on 
Liver  Cell  Release 


Standard  tissue  cylinders  were  obtained  from 
the  livers  of  normal  rats.    These  were  incubated 
for  20  minutes  at  37  °C,  in  either  10  percent 
necrotic  extract  in  HBSS,  or  in  BSA  control  solu- 
tions.   The  cylinders  were  shaken  for  either  10  or 
40  minutes.    Representative  results  of  4  separate 
experiments  given  in  table  5  show  that  after  10 
minutes  shaking,  3  times  greater  volume  of  cells 
were  released  from  liver  cylinders  pretreated  with 
necrotic  extract  than  from  the  control  series; 
after  40  minutes  shaking,  the  ratios  of  extract- 
treated  to  controls  increased  to  50:1. 

The  results  of  experiments  (A)  through  (D)  show 
clear-cut  differences  in  the  mechanical  proper- 
ties of  different  parts  of  the  same  tumor,  and  in 
the  tissues  surrounding  them.    The  changes  appear 


to  be  related  to  the  presence  and  diffusion  of 
necrotic  material.    If  the  increased  detachment 
patterns  is  due  to  liberated  lysosomal  enzymes  in 
which  the  necrotic  material  is  rich,  then  it  must 
be  remembered  that  some  of  these  may  be  contribut- 
ed by  macrophages  and  polymorphs  [16]  which  are 
frequently  found  in  association  with  tumors  and 
necrotic  tissues. 

5.  Conclusions 

If  these  results,  based  on  cell  detachment,  are 
indicative  of  changes  detectable  by  ultrasound, 
and  studies  presently  underway  will  hopefully 
clarify  this,  then  three  positive  suggestions  come 
from  this  work: 

A.  Attention  must  be  given  to  the  microen- 
vironmental  conditions  under  which  specimens  are 
maintained  while  being  examined,  and  the  time 
spent  on  examination  must  be  minimized. 

B.  It  is  not  enough  to  determine  the  charac- 
teristics of  whole  tumors;  individual  regions 
must  be  characterized  and  identified  at  a  "micro- 
scopic level". 

C.  Changes  consequent  to  degeneration,  occur- 
ring in  a  cancer,  and  in  the  nonmalignant  tissues 
surrounding  it,  may  enhance  its  detection  by  ultra- 
sound by  creating  new  differentials  within  these 
tissues. 

Acknowledgments 

My  thanks  are  due  to  Dr.  H.  Bicher  and  Mr.  L. 
D'Agostino,  Department  of  Radiotherapy  for  making 
measurements  of  oxygen-tension,  and  to  Ms.  J. 
Holmes,  D.  Lombardo  and  Mr.  D.  Graham  for  their 
technical  assistance. 

This  work  was  partially  supported  by  Grants 
#PDT-14  from  the  American  Cancer  Society  Inc. 
and  CA-17609  from  the  National  Institutes  of 
Health. 


Table  5. 

Release  of  parenchymal  cells 

from  normal 

liver  ±  nectotic 

extract. 

Experimental  (E) 

Controls  (C) 

Time 

Numbers 

Clump 

Numbers 

Clump 

Volume 

shaken 

released 

diameter 

rel eased 

diameter 

""atio 

(X  1000) 

(pm) 

(X  1000) 

(ym) 

E:C 

±  SE  (n) 

±  SE  (n) 

±  SE  (n) 

±  SE  (n) 

10  m 

88  ±  4.4 

31  +  1.2 

60  ±  3.5 

19  ±  0.9 

3:1 

(10) 

(137) 

(9) 

(127) 

40  m 

250  ±  4.9 

51.3  ±  2.9 

40  ±  4.8 

25.6  ±  1.6 

50:1 

(10) 

(100) 

(10) 

(100) 

References 

[1]    Weiss,  L. ,  The  Cell  Periphery,  Metastasis 
and  other  Contact  Phenomena  (N.  Holland 
Press,  Amsterdam,  1967). 

[2]    Weiss,  L. ,  ed..  Fundamental  Aspects  of 

Metastasis  (North-Holland/American  Elsevier, 
19761:: 

[3]    Weiss,  L.,  Cell  detachment  and  metastasis, 
Gann  (1977a)  (in  press). 


[4]    Weiss,  L.,  Tumor  necrosis  and  cell  detachment, 
Internat.  J.  Cancer  (1977b)  (in  press). 

[5]    Altman,  P.  L.  and  Katz,  D.  D.,  eds..  Cell 

Biology,  p.  61  et  seq. ,  F.A.S.E.B.,  Bethesda, 
Maryland  (1976). 

[6]    Warburg,  0.,  Ober  den  Stattwechsel  der 
Tumoren  (Springer,  Berlin,  1956). 

[7]    McLimans,  W.  P.,  Crouse,  E.  J.,  Tunnah, 
K.  B.,  and  Moore,  G.  E.,  Kinetics  of  gas 


321 


diffusion  in  mammalian  cell  culture  systems. 
1.    Experimental,  Biotech.  Bioeng.  10,  725- 
740  (1968). 

[8]    Cater,  D.  B.,  Silver,  I.  A.,  and  Wilson, 
G.  M. ,  Apparatus  and  technique  for  the 
quantitative  measurement  of  oxygen  tension 
in  living  tissues,  Proc.  Roy.  Soc.  Lond. 
(Series  B)  151,  256-276  (1959/1960): 

[9]    Gornandt,  L.  and  Kessler,  M. ,  PO2  Histo- 
grams in  Regenerating  Liver  Tissue,  in 
Oxygen  in  Tissues,  M.  Kessler,  D.  F.  Birnley, 
L.  C.  Clark,  D.  W.  Lubbers,  I.  A.  Silver, 
and  J.  Strauss,  eds.,  p.  288  (Urban  & 
Schwarzenberg,  Miinchen-Berl in-Wien ,  1973). 

[10]    Ericsson,  J.  L.  E.,  Mechanism  of  Cellular 
Autophagy,  in  Lysosomes  in  Biology  and 
Pathology,  J.  T.  Dingle  and  H.  B.  Fell, 
eds..  Vol.  2,  pp.  345-394  (North-Holland, 
Amsterdam,  1969). 

[11]    Fell,  H.  B.  and  Weiss,  L.,  The  effect  of 
antiserum,  alone  and  with  hydrocortisone, 
on  foetal  mouse  bones  in  culture,  J.  Exptl. 
Med,  121,  551-560  (1965). 


[12]    Weiss,  L. ,  Studies  on  cellular  adhesion  in 
tissue  culture.    VIII.    Some  effects  of 
antisera  on  cell  detachment.    Exptl.  Cell 
Res.  37,  540-551  (1965). 

[13]    Dunn,  F. ,  Ultrasonic  Absorption  by  Bio- 
logical Materials,  in  Ultrasonic  Energy, 
E.  Kelly,  ed. ,  pp.  51-65  (Univ.  of  Illinois 
Press,  Urbana,  1965). 

[14]    Slack,  N.  H.,  Blumenson,  L.  E.,  and  Bross, 
I.  D.  J.,  Therapeutic  implications  from  a 
mathematical  model  characterizing  the  course 
of  breast  cancer.  Cancer  24,  960-971  (1969). 

[15]    Blumenson,  L.  E.  (1972)  Cited  by  L.  Weiss, 
Some  Patho-biological  Considerations  of 
Breast  Cancer  by  Ultrasonic  Holography,  in 
Ultrasonic  imaging  and  holography,  G.  W. 
Stroke,  W.  E.  Kock,  Y.  Kikuchi,  and  J. 
Tsujinchi,  pp.  567-585  (Plenum  Press,  New 
York,  1974). 

[16]    Davies,  P.  and  Allison,  A.  C,  The  Secre- 
tion of  Lysosomal  Enzymes,  in  Lysosomes  in 
Biology  and  Pathology,  J.  T.  Dinqle  and 
R.  T.  Dean,  eds..  Vol.  5,  pp.  61-98  (North- 
Holland,  Amsterdam,  1976). 


322 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979) 


A  HUMAN  ABDOMINAL  TISSUE  PHANTOM 


P.  D.  Edmonds,  Z.  Reyes,  D.  B.  Parkinson 

Stanford  Research  Institute  International 
Menlo  Park,  California    94025,  U.S.A. 

R.  A.  Filly  ; 

University  of  California  Medical  Center 
San  Francisco,  California    94122,  U.S.A. 

H.  Busey 

Picker  Corporation 
Northford,  Connecticut    06472,  U.S.A. 


The  objective  of  this  work  was  to  determine  the  feasibility  of  constructing  a 
phantom  that  would  simulate  human  abdominal  tissues  when  interrogated  by  advanced 
B-scan  diagnostic  ultrasound  equipment  operating  at  2.25  and  3.5  MHz.  Satisfactory 
results  were  obtained  with  gelatin  based  components  having  dispersed  scatterers 
and  embedded  sponge,  plastic  tubes  and  rubber  bulbs  filled  with  saline  solution. 
Stability  of  the  gelatin  was  achieved  by  addition  of  a  stabilizing  agent,  a  pre- 
servative and  impervious  coatings  to  prevent  water  evaporation. 

Key  words:    Gelatin;  human  abdominal  tissues;  phantom. 


In  the  course  of  producing  and  installing  ultra- 
sonic diagnostic  equipment,  a  need  arises  for  a 
phantom  object  with  stable  acoustic  properties, 
that  will  generate  displays  suitable  for  optimizing 
the  adjustment  of  various  pre-set  equipment  levels, 
particularly  in  the  grey-scale  circuitry.    The  need 
extends  beyond  the  concerns  of  a  single  company  be- 
cause there  is  a  lack  of  basic  standardization  in 
the  industry.    The  general  availability  of  a  suit- 
able phantom  would  improve  uniformity  of  displays 
obtained  with  different  ultrasonic  imaging  systems. 

This  paper  describes  the  evolution  of  a  phantom 
object  intended  to  simulate  many  of  the  ultrasonic 
properties  of  human  abdominal  tissue  when  imaged 
by  a  Picker  Echoview  System  801  Laminograph. 
Figure  1  is  an  abdominal  B-scan  showing  the  variety 
of  textures  to  be  simulated. 

The  design  specification  called  for: 

•  attenuation  coefficient  at  2.25  MHz  = 
1.7  ±  0.1  dB/cm; 

•  velocity  of  propagation  at  22  °C  (72  °F)  = 
1540  ±  50  m/s; 

•  simulation  of  fine  and  coarse  texture  of 
normal  and  pathological  liver,  as  well  as  the 
specular  reflection  of  capsular  interfaces 
and  blood  vessels,  cysts  with  or  without 
sediment,  and  abdominal  aorta,  utilizing  the 
full  range  of  grey  levels; 

•  density  =  1.05  +0.1  g/cm^; 

•  mechanical  stability  up  to  65  °C  (150  ^F); 

•  flexible  exterior  surface  permitting  limited 
sector  scanning; 

•  abrasion  resistant  surface  in  the  presence 
of  mineral  oil  and  aqueous  gels. 


Fig.  1.    Human  abdominal  B-scan  showing  variety 
of  textures  to  be  simulated. 


Less  importance  was  attached  to  reproducing  spe- 
cific organ  geometries  or  spatial  relationships. 

Initial  work  focused  on  the  selection  of  a 
base  material.    Classes  of  materials  considered 
were  room  temperature  vulcanizing  (RTV)  silicone 
rubbers,  "pc"  rubbers,  gelatin  gels,  3M®  percus- 
sion pads  and  "Hydron®"  polymer.    In  order  to 
simulate  "texture",  it  would  be  necessary  to  in- 
corporate scattering  centers  in  the  base  material 
Hence,  some  part  of  the  overall  attenuation  co- 
efficient, A,  of  an  acceptable  simulating  mate- 
rial would  be  attributable  to  scattering  and  the 
residual  absorption  coefficient,  a,  of  the  base 


323 


material  without  scattering  had  to  be  correspond- 
ingly less  than  the  desired  overall  attenuation 
coefficient:    a  =  A  -  a^,  where       is  the  at- 
tenuation per  centimeter  due  to  scattering. 

Measured  absorption  coefficients  and  propaga- 
tion velocities  of  the  candidate  base  materials 
are  given  in  table  1.    The  absorption  data  were 
obtained  by  the  substitution  method  and  the 
velocity  data  by  the  pulse  travel  time  method 
[1]^.    The  techniques  of  sample  preparation  as 

Table  1.    Measured  material  properties 
at  2.25  MHz,  22  °C. 


Material 


Absorption 
coefficient 
(dB  cm-i) 


Propagation 
velocity 


Notes 


General  Electric 

RTV  602 
RTV  515 

Goodrich 

#35080  "pc" 
rubber 


3M®  percussion 
pad,  style  3290 

"Hydron®  " ,  dry 
"Hydron®  " ,  wet 

10%  gelatin/water 
+  1%  chrome  alum 

Specification 


1 .34  ±  0.2 
2.55  ±  0.3 


2.84  ±  0.11 

1.7    ±  0.2 

16  ±  2 
13  ±  2 

<0.5 
<0.1 

<1.7    ±  0.1 


980  ±  10 
980  ±  10 


1546  ±  3 

1490  ±  10 

2050  ±  12 
1950  ±  12 

1539  ±  10 

1540  ±  50 


0.25%  SRC05  catalyst;  cured  48  hours  at  room 
■temperature. 
10%  615B  catalyst;  cured  48  hours  at  room 
temperature. 

Sample  formed  from  uncured  rubber  stock  by  milling 
,and  molding  for  45  minutes  at  6.6  kg/m^  and  150  °C. 
Measured  as  received  in  sheet  form. 
^The  sample  from  Hydron,  Inc.    (2-hydroxy  ethyl 
methacrylate  polymer  in  water)  was  dried  and  then 
molded  at  175  °C  under  pressure.    This  produced  a 
hard,  brittle,  opaque  cylinder  which  was  immersed 
in  distilled  water  to  swell  and  soften.    The  re- 
sultant swelling  was  irregular.    Internal  bubbles 
and  interfaces  became  visible.    The  attenuation 
was  greater  than  10  dB/cm  at  2.25  MHz,  22  °C. 
This  sample  was  then  recovered  by  dissolving  in 
ethanol ;  the  solution  was  placed  in  a  dessicator 
in  the  presence  of  ethanol  and  dried  over  a  period 
of  2  months  to  yield  a  translucent  1/8  inch  thick 
disc.    The  "dry"  sample  was  this  disc  equilibrated 
j,with  air. 

A  portion  of  the  disc  was  cut  and  immersed  in  dis- 
tilled water  for  24  hours.    It  became  opaque  white 
and  evidently  different  from  contact  lens  material. 
This  was  the  "wet"  sample. 

^Photographic  gelatin  swollen  in  water  and  softened 
by  heating.    Diluted  to  10%  solution  in  distilled 
water;  0.1%  sodium  benzoate  added  as  a  preserva- 

,tive.    Softening  point  25  °C  (77  °F). 
1%  chrome  alum  added  as  a  stabilizer.  Softening 
point  >  95       (203  °F). 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


described  in  the  notes  may  have  influenced  the 
data.    It  is  evident  that  only  RTV  602  and  gela- 
tin had  sufficiently  low  absorption  coefficients 
to  permit  additional  attenuation  due  to  scatter- 
ing without  exceeding  the  specified  overall  at- 
tenuation coefficient.    Both  these  materials  had 
disadvantages.    The  velocity  of  RTV  compounds  is 
only  two  thirds  of  the  desired  figure.    The  gela- 
tin is  liable  to  drastic  changes  of  dimensions 
and  properties  as  water  evaporates  from  the  gel. 

Consideration  was  given  to  accepting  a  base 
material  (such  as  RTV)  with  a  velocity  outside 
the  specified  range,  and  scaling  the  sizes  of 
phantom  components  so  that  the  travel  times  of 
pulsed  signals  would  correspond  to  those  in  the 
organs  being  simulated.    However,  this  approach 
was  judged  inherently  unacceptable  and  it  was 
anticipated  that  a  base  material  could  be  found 
or  formulated  with  the  desired  combination  of 
properties.    Also  the  RTV  silicone  rubber  was  be- 
lieved to  suffer  from  long  term  instability  of 
acoustic  properties  [2].    In  retrospect,  it 
seems  probable  that  the  same  precautions  to  com- 
bat instability  that  are  described  below  for 
gelatin-based  compounds  would  be  applicable  to 
RTV  silicones  in  applications  where  scaling  sizes 
might  be  acceptable. 

Gelatin  gels  were  considered  suitable  for  sub- 
sidiary experiments  in  which  various  scattering 
materials  were  dispersed  in  the  gelatin  before 
molding;  the  objective  was  to  produce  a  fine 
textured  image  and  the  specified  attenuation  co- 
efficient.   From  both  laboratory  measurements 
and  imaging,  with  the  controls  on  the  diagnostic 
equipment  set  for  abdominal  imaging,  it  was  found 
that  a  1  percent  dispersion  of  a  cellulose  fiber 
or  a  0.05  percent  dispersion  of  glass  micro- 
spheres in  10  percent  gelatin  in  water  produced 
the  desired  results.    The  cellulose  fiber  had  an 
average  fiber  length  of  70  to  140  ym;  the  glass 
microspheres  ranged  from  10  to  250  \m  in  diam- 
eter. 

The  next  objective  was  to  identify  materials 
which,  when  embedded  in  gelatin,  would  simulate 
the  acoustic  aspects  of  abdominal  organs.  Trial 
and  evaluation  led  to  a  selection  of  materials 
each  of  which  was  embedded  separately  in  a  gela- 
tin block  containing  scatterers.    These  blocks 
could  then  be  stacked  in  various  sequences  and 
coupled  with  mineral  oil  for  trial  imaging. 

Figure  2  illustrates  the  various  inclusions 
imaged  at  3.5  MHz.    The  top  block  contains  the 
base  scattering  material  consisting  of  1  percent 
cellulose  fiber  in  10  percent  gelatin  in  water. 
A  gelatin  filled  natural  sponge  is  placed  near 
the  left  edge. 

The  middle  layer  contains  a  soft  rubber  bulb 
filled  with  saline  solution  to  simulate  a  large 
cyst  or  fluid  filled  organ.    The  bright  spot  in 
the  center  of  the  bulb  is  an  artifact  caused  ap- 
parently by  a  focusing  action.    The  right  side 
of  the  center  layer  contains  a  natural  sponge 
filled  with  the  gelatin  containing  scatterers, 
which  might  be  used  to  simulate  metastatic  liver 
disease. 

The  lowest  level  contains  a  variety  of  tubing. 
The  parallel  lines  display  the  1/8  inch  wall  of  a 
soft  rubber  tube  filled  with  clear  gelatin.  The 
upper  wall  is  seen  near  the  junction  between  the 
blocks  and  the  lower  wall  in  the  center  of  the 
block.    This  is  offered  as  an  example  of  the 
resolution  available  at  a  depth  of  14  cm  through 


324 


Fig.  2.    Phantom  components  imaged  at  3.5  MHz. 


the  various  phantom  components.    At  the  lower 
right,  some  cross  sectional  images  of  tubes  com- 
pare favorably  with  normal  images  of  the  portal 
vein  or  abdominal  aorta.    The  tubes  are  poly- 
olefine  shrink  tubing  5/16  inch  o.d.  with  0.02 
inch  wall  and  clear  vinyl  tubing  7/16  inch  diam- 
eter with  1/16  inch  wall.    These  phantom  com- 
ponents were  scanned  at  both  2.25  MHz  and  3.5 
MHz  with  improvements  in  resolution  at  the  higher 
frequency  corresponding  to  that  found  in  bio- 
logical tissue.    Figure  3  shows  an  image  at  2.25 
MHz  with  resolution  clearly  inferior  to  that  of 
figure  2. 


Fig.  3.    Phantom  components  imaged  at  2.25  MHz. 


The  echo  discontinuities  and  interfaces  be- 
tween layers  should  be  eliminated  when  a  con- 
tinuously molded  phantom  is  produced  rather 
than  the  individual  blocks  shown  here. 

Gelatin  gels  were  originally  considered  un- 
desirable as  the  base  material  because  of  their 
obvious  instability  under  heat  and  evaporation 
of  water  and  their  susceptibility  to  degradation 
while  nourishing  mold  organisms.    However,  the 
addition  of  chrome  alum  as  a  thermal  stabilizer 
and  sodium  benzoate  as  a  preservative  eliminated 
two  of  these  undesirable  features.    The  problem 
of  evaporation  of  water  is  not  so  easily  solved 


if  it  is  required  that  the  propagation  velocity 
remain  in  the  range  1540  ±  50  m/s.    Two  solutions 
appear  possible:    a  substantial  fraction  of  the 
water  can  be  replaced  by  hydrophilic  compounds 
that  have  at  least  as  high  a  boiling  point  as 
water  and  preferably  higher.    Table  2  shows 
some  results  of  this  approach;  all  the  velocities 
are  too  high.    Alternatively,  samples  can  be 
coated  to  prevent  the  water  from  evaporating; 
this  approach  was  undertaken. 


Table  2. 

Velocities  at 

2.25  MHz,  22 

°C. 

Composition 

Velocity 

(m/s) 

10%  gelatin,  89%  water 
1%  chrome  alum 

1539  ± 

10 

10%  gelatin, 
70%  glycerol 

20%  water 

1900 

10%  gelatin, 
67%  glycerol 

20%  water 
3%  PEG  200 

1900 

10%  gelatin, 
45%  PEG  600 

45%  water 

1770 

10%  gelatin, 
60%  PEOa  200, 

20%  water 
10%  sorbitol 

1749 

10%  gelatin, 
45%  PEG  200 

45%  water 

1 746 

10%  gelatin, 
45%  DEG 

45%  water 

1717 

10%  gelatin, 
70%  DEGb 

20%  water 

1703 

10%  gelatin, 
60%  carbitol 

30%  v-;ater 

1644 

^PEG  =  Polyethylene  glycol 
DEG  =  Diethylene  glycol 

First  the  base  material  was  slightly  modified 
by  inclusion  of  5  percent  polyethylene  glycol  200 
in  the  formulation,  to  provide  compatibility  with 
the  first  of  five  coatings  that  were  applied  to 
the  test  blocks  of  gelatin.    Then,  successive 
coatings  of  a  butadiene-acrylic  rubber,  a  mixture 
of  this  rubber  with  paraffin,  and  two  coats  of  a 
mixture  of  the  same  rubber  with  polyvinyl  acetate 
were  applied.    By  grading  the  coatings  in  this 
manner,  good  adhesion  and  acoustic  contact  be- 
tween coatings  was  obtained,  while  providing  the 
desired  properties  in  succession.    The  paraffin 
prevented  water  evaporation,  and  the  polyvinyl 
acetate  provided  the  tough  flexible  exterior  sur- 
face.   Water  evaporation  through  this  coated 
specimen  has  been  reduced  to  0.25  percent  per  day 
with  the  specimen  open  to  air.    It  is  expected 
that  a  practical  phantom  would  be  enclosed  in  a 
protective  case,  within  which  the  humidity  can 
be  somewhat  controlled  as  long  as  a  reservoir  of 
water  is  replenished.    Consequently,  the  problem 
of  water  evaporation  is  considered  to  be  under 
control,  although  further  improvements  in  coating 
formulations  are  under  study. 

In  conclusion,  specific  combinations  of  mate- 
rials have  been  identified  to  make  possible  the 
construction  of  a  phantom  object  to  simulate 
many  of  the  acoustic  characteristics  of  human 
abdominal  tissue.    Such  a  phantom  should  aid  in 
the  standardization  of  operating  conditions  of 
ultrasonic  diagnostic  equipment. 


325 


Acknowledgments 


References 


The  authors  wish  to  acknowledge  the  invaluable 
assistance  of  William  Mullen  in  performing  acous- 
tical measurements,  Marie  Comas  and  Irving  Illing 
in  preparing  test  samples,  and  Linda  McKay,  Cheryl 
Wilson  and  Diane  Eskelson  in  scanning  the  many 
test  components. 

The  work  reported  in  this  paper  was  carried  out 
on  behalf  of  the  Picker  Corporation. 


[1]    McSkimin,  H.  J.,  Ultrasonic  Methods  for 
Measurement,  in  Physical  Acoustics,  W.  P. 
Mason,  ed.  ,  Vol.  lA,  pp.  271-334  (Van 
Nostrand,  New  York,  1965). 

l"2]    Private  communication  at  WFUMB  Workshop, 
August  1976. 


326 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed. ,  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


TISSUE  SIMULATORS  FOR  DIAGNOSTIC  ULTRASOUND 


Reginald  C.  Eggleton  and  James  A.  Whitcomb 

Ultrasound  Research  Laboratories  of 
The  Indianapolis  Center  for  Advanced  Research,  Inc. 
Indianapolis,  Indiana    46202,  U.S.A. 


Factors  are  presented  related  to  the  design,  construction  and  use  of  phantoms  to 
replace  human  patients  or  subjects  in  the  development,  testing,  clinical  training  and 
promotion  of  ultrasonic  diagnostic  equipment.    Materials  are  described  which  match 
the  acoustic  properties  of  tissue.    Realistic  echograms  can  be  obtained  by  scanning 
properly  configured  phantoms  using  such  materials. 

The  phantoms  may  be  simple  geometric  test  patterns,  sections  of  human  torso  or 
complete  human  torsos  simulating  dynamic  cardiovascular  and  respiratory  movements  to 
evaluate  the  real-time  systems. 


Key  words:    Training  phantoms;  tissue  signature;  tissue  simulators;  ultrasonic 
phantoms . 


1.  Introduction 

Tissue  simulators  or  phantoms,  as  they  are 
sometimes  called,  may  be  utilized  in  the  field 
of  medical  ultrasound  for  purposes  of  quantita- 
tively and  qualitatively  evaluating  the  perfor- 
mance of  diagnostic  or  therapeutic  equipment. 
They  may  also  be  used  for  purposes  of  training. 
Yet  another  application  is  basic  study  of  the 
interaction  of  sound  with  tissues  for  purposes 
of  improving  existing  diagnostic  criteria,  under- 
standing the  mechanisms  of  acoustic  image  forma- 
tion and  for  devising  new  diagnostic  methods. 

The  design  and  construction  of  realistic 
ultrasonic  phantoms  to  substitute  for  human  pa- 
tients or  subjects  during  the  early  training  of 
technicians  and  physicians  is  based  upon  knowledge 
from  several  diverse  fields  of  study.    The  anatomy 
of  the  phantom  must  be  correct  with  respect  to  the 
details  visualized  by  ultrasound.    Landmark  fea- 
tures used  by  ul trasonographers  to  orient  and 
identify  the  scan  planes  must  be  accurately  rep- 
resented.   The  acoustic  properties  of  biological 
materials  important  to  the  appearance  of  the 
image  must  be  catalogued  for  the  tissues  that  will 
be  represented  in  the  phantoms.    A  knowledge  of 
materials  that  can  be  used  in  the  construction  of 
the  phantoms  is  necessary.    It  is  essential  to 
know  how  to  control  various  acoustic  properties 
such  as  attenuation,  scattering,  speed  of  sound, 
impedance,  etc.    Likewise  it  is  necessary  to  have 
a  knowledge  of  the  chemistry  of  the  materials  to 
select  components  which  will  give  the  phantoms 
adequate  stability  and  a  useful  life. 

A  knowledge  of  fabricating  techniques  required 
in  the  construction  of  the  phantoms  is  also  neces- 
sary.   The  methods  for  casting  and  welding  of 
phantom  materials  to  obtain  the  shapes  and  con- 
tours necessary  to  adequately  model  the  human 


anatomy  are  also  required.    It  is  also  desirable 
to  have  the  necessary  artistic  ability  to  create 
an  aesthetically  acceptable  model  which  will  have 
the  feel  and  appearance  of  the  structures  it  pur- 
ports to  represent. 

2.    Materials  and  Methods 

Hydrocolloid  gels  are  ideally  suited  for  con- 
struction of  ultrasound  tissue  phantoms  because  of 
the  flexibility  and  control  over  acoustic  param- 
eters   and  the  stability  with  respect  to  time. 
There  are  two  hydrocolloid  systems  which  are  par- 
ticularly suitable  for  this  application;  one  which 
is  reversible,  and  the  other  irreversible.    A  com- 
bination of  the  two  allows  great  flexibility  in 
fabricating  techniques. 

Irreversible  Hydrocolloid  -  Alginic  acid  was 
first  prepared  and  analyzed  by  Stanford  in  1886 
[1]^  who  established  it  as  a  weak  organic  acid 
that  readily  forms  salts  and  bases.    Nelson  and 
Cretcher  [2]  hydrolyzed  alginic  acid  to  obtain  the 
salt  of  D-mannuronic  acid.    X-ray  and  chemical 
evidence  suggest  that  alginic  acid  is  a  high  molec- 
ular weight  polymer  made  up  of  D-mannuronic  acid. 
Several  different  methods  indicate  that  the  com- 
mercial sodium  alginates  have  a  molecular  weight 
between  3200  and  200,000  and  a  degree  of  polymeri- 
zation from  0.180  to  0.930.    Recent  evidence  has 
been  obtained  showing  that  the  polymer  molecule  is 
entirely  linear,  consisting  of  linked  mannuronic 
acid  units  and  a!?.(l-^4)  linked  glutamic  acid  units 
[3-5].    Models  constructed  from  x-ray  diffraction 
data  indicate  that  the  D-mannuronic  acid  units  are 
in  the  C-1  chair  configuration,  whereas  the  t-glu- 
tamic  acid  units  are  in  the  1-C  configuration  [6]; 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


327 


thus  algin  is  a  copolymer  and  not  simply  a  mix- 
ture of  mannuronic  acid  and  gluronan. 


1/1 


OH 
I 

c=o 
I 

c  0 


OH 

,1  

I  I 

H  H 


H 
I 

-C 
N  H 

°M 

c 


c=o 
I 

c  


H  /I 


\/\r  /\/v 


OH/ 
1/ 


The  copolymers  are  linked  together  with  an 
alternating  composition.    The  statistical  descrip- 
tion of  the  structure  of  algin  is  best  obtained 
by  assuming  that  the  copolymer  is  formed  accord- 
ing to  the  penultimate-unit  theory  of  addition 
copolymerization  [7]. 

Water  soluble  salts  of  alginic  acid  include 
those  of  alkaline  metals,  ammonia  and  low  molec- 
ular weight  amines.    Sodium  alginate  is  the  most 
common  salt  and  is  readily  available.  Solutions 
can  be  made  in  either  hot  or  cold  water.  Algin- 
ates dissolve  most  easily  when  sifted  into  water 
and  vigorously  agitated.    High  speed  stirring  and 
gradual  addition  of  solid  increases  the  rate  of 
dissolution.    Facilitation  of  the  process  can  be 
achieved  by  wetting  the  alginate  with  ethanol  or 
glycerol  or  other  water-miscible  liquid  before  it 
is  added  to  the  water.    With  a  good  dispersion 
the  alginate  dissolution  should  be  completed  in 
a  few  minutes. 

The  properties  of  alginate  solutions  can  be 
changed  readily  to  provide  either  slow,  fast  or 
intermediate  types  of  flow  [8].    Algin  solutions 
are  strongly  influenced  by  covalent  ions  such  as 
calcium  which  increase  the  viscosity  and  lower  the 
flow  properties,  with  little  change  in  the  acous- 
tic properties.    Alginates  show  increasing  non- 
Newtonian  flow  characteristics  with  increased  de- 
gree of  polymerization  and  substitution  of  calcium 
for  sodium  [9],  and  become  thixotropic  at  low 
levels  of  calcium.    The  effect  of  this  ion  in 
causing  increased  viscosity  is  more  pronounced  in 
alginates  with  higher  D-mannuronic  acid  content 
[10].    The  viscosity  of  sodium  alginate  solutions 
can  be  depressed  by  using  sodium  with  potassium 
salts.    The  high  viscosity  alginate  can  attain  a 
viscosity  of  at  least  2000  cps  at  a  concentra- 
tion of  1  percent  in  water,  whereas  a  low  viscosi- 
ty product  will  have  a  viscosity  of  less  than  10 
cps  at  the  same  concentration  as  a  function  of 
molecular  weight.    Algin  solutions  behave  like 
other  fluids  in  their  dependence  of  viscosity  on 
temperature  and  may  decrease  approximately  2.5 
percent  per  °C  with  increasing  temperature.  The 
viscosity  of  water  soluble  algin  salts  changes 
only  slightly  as  a  function  of  changes  in  pH,  in 
the  range  of  4  to  10. 

The  cross-linking  created  by  the  substitution 
of  calcium  for  sodium  renders  the  alginate  in- 
soluble.   When  the  insoluble  salt  is  formed  by 
the  reaction  of  the  sodium  alginate  in  solution 
with  a  calcium  salt,  for  example,  the  calcium  ion 
may  replace  the  sodium  ions  in  two  adjacent  mole- 
cules to  produce  a  cross-linking  between  the  two 
molecules.    As  the  reaction  progresses,  a  cross- 
linked  molecular  complex  or  polymer  network  forms. 
Such  a  network  constitutes  the  micelles  structure 
of  the  gel.    Solutions  of  sodium  alginate  are 
stable  for  long  periods  at  room  temperature  but. 


like  other  polysaccharides,  are  depolymerized  in 
the  presence  of  a  number  of  autoxidizable  sub- 
stances such  as  phenolic  compounds.    Sodium  algin- 
ates tend  to  retard  the  growth  of  many  microorgan- 
isms or  are  passive  with  respect  to  others.  Al- 
though the  algin  is  resistant  to  the  attack  of 
microorganisms,  its  solutions  are  subject  to  some 
bacterial  action  during  prolonged  storage.  Agents 
such  as  Zepharin  chloride  can  be  added  in  small 
quantities  to  further  decrease  the  bacterial 
growth.    Small  amounts  of  calcium  increase  the 
stability  of  sodium  alginate  solutions. 

Pure  algins  in  water  form  gels  having  excellent 
optical  clarity.    By  altering  the  formulation, 
gels  can  be  varied  in  texture  from  those  which  are 
soft  and  exhibit  flow  characteristics  to  those 
which  are  tough  and  elastic.    This  range  of  prop- 
erties exhibits  little  effect  on  the  acoustic  im- 
pedance or  acoustic  velocity  characteristics  of 
the  material.    As  a  result,  algin  gels  are  ideal- 
ly suited  to  tissue  simulation  in  that  structures 
can  be  molded  and  the  attenuation  and  scattering 
properties  can  be  independently  controlled. 

In  general,  the  gels  are  formed  by  the  gradual 
and  uniform  release  of  either  calcium  or  hydrogen 
ions,  or  a  combination  of  the  two  throughout  the 
algin  solution.    The  setting  time  can  be  control- 
led by  the  addition  of  a  limited  amount  of  a  com- 
pound such  as  phosphate  or  polyphosphate  that  is 
capable  of  combining  with  calcium.  Frequently 
calcium  sulphate  and  a  phosphate  buffer  are  com- 
bined to  control  the  gelation  rate  of  the  algin- 
ate solutions. 

The  acoustic  properties  of  the  gel  are  deter- 
mined largely  by  the  water.    Two  percent  sodium 
alginate  solution  is  the  usual  concentration  in 
combination  with  small  quantities  of  salts.  The 
acoustic  velocity  using  the  phase  method  [13]  is 
approximately  1550  meters  per  second  at  37  °C; 
the  density  of  the  2  percent  gel  is  approximately 
1.04  grams  per  cubic  centimeter,  thus  the  acous- 
tic impedance  falls  in  the  center  of  the  plot  of 
tissue  characteristics  (see  figure  5).    The  at- 
tenuation for  the  gel  without  the  scatterers  in- 
corporated is  approximately  0.5  dB  cm"^  MHz"i. 

Reversible  Hydrocolloid  -  An  organic  hydro- 
philic  colloid  of  polysaccharide  can  also  be  uti- 
lized for  construction  of  acoustic  test  objects. 
A  commonly  available  form  utilizes  the  sulphuric 
ester  of  a  linear  polymer  of  galactose.  This 
material  undergoes  a  reversible  reaction,  sol  t 
gel,  under  the  influence  of  temperature  changes. 

In  the  gel  state  the  binding  forces  are  second- 
ary forces,  e.g.,  dipole  interactions  or  van  der 
Waals  forces.    When  the  concentration  of  the  dis- 
persed phase  in  the  hydrocolloid  is  of  the  proper 
amount,  the  sol  can  be  changed  to  a  semisolid  gel 
when  the  temperature  is  decreased  below  the  criti- 
cal point  (^^  110°F).    The  gelation  is  complete  at 
approximately  100°F.    In  this  process  the  dis- 
persed phase  agglomerates  to  form  chain  fibrils, 
referred  to  as  "micelles."    The  process  continues 
until  a  three-dimensional  matrix  is  formed  with 
much  branching  and  intermeshing  of  the  micelles 
to  form  a  "brush  heap"  structure.    The  intermesh- 
ing and  weak  binding  forces  contribute  strength 
to  the  structure.    The  interstices  of  this  struc- 
ture are  filled  with  water;  the  water  predominant- 
ly contributes  to  the  acoustic  properties  of  the 
gel.    It  will  be  noted  upon  warming  that  the  sol 
st-ate  is  again  attained  but  the  temperature  must 
be  maintained  at  120°F  for  a  period  of  time  be- 


328 


fore  liquefaction  is  achieved.    The  temperature 
lag  between  gelation  and  liquefaction  is  described 
as  a  hysteresis  phenomenon. 

In  the  gel  state  the  structure  exhibits  both 
elastic  and  viscous  properties.    If  the  gel  is  de- 
formed it  requires  time  to  recover.    Likev/ise,  if 
the  gel  loses  water,  it  will  tend  to  absorb  water 
approaching  the  initial  state.    The  strength  of 
the  gel  is  decreased  at  elevated  temperatures, 
but  can  be  substantially  increased  with  the  addi- 
tion of  a  small  amount  of  borax.    The  borax  ac- 
complishes this  by  increasing  the  interaction  of 
the  micelle  framev^ork.    The  borax  also  increases 
the  viscosity  in  the  sol  state.    The  addition  of 
absorbers  and  scatterers  also  increases  the  vis- 
cosity, as  well  as  adding  to  the  strength  of  the 
gel.    A  suitable  composition  would  consist  of 
v/ater  (83.5  percent),  polysaccharide  (14.2  per- 
cent), potassium  sulphate  (2  percent)  and  borax 
(0.2  percent).    The  potassium  sulphate  is  intro- 
duced to  minimize  the  amount  of  syneresis  as  well 
as  imbibition.    The  introduction  of  soluble  salts 
into  the  gel  adjusts  the  osmotic  pressure  of  the 
dispersion  medium  and  thus  helps  to  stabilize  the 
system.    The  polysaccharide  frequently  used  is  ex- 
tracted from  seaweed  and  is  a  linear  polymer  of 
galactose,  shown  below. 


I 

c— 

OH/I 
/  OH 

c 

\ 

 c  — 

I 


H 
I 

c— 

OH/1 
l/OH 


V 

i/\  K 


Fig.  1.    Scattering  test  objects  are  constructed 
with  Dextran  spheres  dispersed  in  a  re- 
versible hydrocolloid  (polysaccharide). 
The  relative  concentration  ranges  from 
0.1  to  100  as  marked  to  produce  a  range 
of  scattering  intensities.    To  scan  the 
test  object  the  transducer  is  placed 
against  a  thin  plastic  window  at  the  top 
of  the  device. 


The  acoustic  properties  of  the  gel  with  embed- 
ded scatterers  and  absorbers  closely  match  those 
of  tissue.    The  specific  acoustic  properties  can 
be  adjusted  by  varying  the  composition  of  the  gel. 

The  choice  of  whether  to  use  the  reversible  or 
irreversible  hydrocolloid  will  be  dictated  by  the 
fabrication  requirements.    For  example,  an  inclu- 
sion within  a  simulated  organ  could  be  construct- 
ed from  an  irreversible  hydrocolloid  and  then 
later  cast  in  the  reversible  hydrocolloid  so  that 
the  temperature  of  the  embedding  material  would 
not  affect  the  shape  of  the  inclusion.    One  obvi- 
ous advantage  of  the  reversible  hydrocolloid  dur- 
ing the  development  phase  is  that  the  materials 
can  be  re-used,  and  it  is  therefore  more  economi- 
cal, but  once  the  design  has  been  established, 
this  advantage  disappears. 

The  acoustic  properties  of  the  reversible  poly- 
saccharide hydrocolloid  gels  are:    density  - 
1.044  g/cm3;  velocity  of  sound  -  1,540  m/s;  at- 
tenuation -  0.4  dB  cm"i  MHz"i.    These  values  are 
remarkably  close  to  those  measured  for  the  ir- 
reversible hydrocolloid  and  for  tissue. 

Test  objects  utilizing  this  material  were  con- 
structed in  which  various  concentrations  of  scat- 
terers were  incorporated  as  layers.    Because  there 
is  no  difference  in  the  characteristic  acoustic 
impedance  across  interfaces,  the  only  reflection 
produced  is  due  to  the  scattering  (no  specular 
reflection  is  obtained).    Figure  1  shows  tv;o  test 
objects  developed  to  display  a  wide  dynamic  range 
of  backscattering  targets.    The  relative  concen- 
tration of  scatterers  is  indicated  for  each  level, 
and  ranges  from  0.1  to  100  as  marked  on  the  test 
object. 

Silastic  Phantoms  -  A  tissue  simulator  not 


previously  described  in  a  publication  was  submit- 
ted to  the  AIUM  Standards  Committee  in  1970  [12]. 
This  silastic  device  contained  both  interfaces  of 
graded  reflectivity  and  scattering  sources.  The 
test  block  v;as  constructed  as  indicated  in  figure 
2  using  silicone  rubber  (RTV  3110)  and  a  silicone 


TEST  BLOCK 


97% 

915 

88% 

8S% 

100% 

0.2 

0.4 

0.6 

0.8 

1.0 

0.0 

100% 

COMPOSITION  OF  TEST  SPECIMEN 


Composition  of  mixture 


Sound  velocity  Characteristic 


Sil icone 

Sil icone 

Density 

at  37.0  °C 

impedance 

rubber 

fluid 

g/cm' 

m/s  X  10-3 

rayl  «  10-5 

RTV  3110 

DE200-20CS 

P 

C 

Z 

100  % 

0  % 

1 .129 

0.960 

1 .084 

97 

3 

1.124 

0.962 

1.082 

94 

6 

1 .120 

0.964 

1.080 

91 

9 

1.115 

0.967 

1.078 

88 

12 

1.110 

0.969 

1.076 

85 

15 

1 .105 

0.971 

1.073 

100 

0.971 

0.975 

0.947 

Fig.  2.    The  tissue  simulator  shown  schematically 
above  was  constructed  using  mixtures  of 
silastic  and  fluid  to  achieve  a  range  of 
impedance  differences  across  an  interface. 


329 


Fig.  3.    Materials  exhibiting  a  range  of  acoustic 
impedances  are  achieved  by  varying  the 
ratio  of  a  two-component  plastic  system 
where  the  individual  components  have  dif- 
ferent acoustic  properties.    In  this  in- 
stance, silicone  rubber  and  silicone  fluid 
are  mixed  to  vary  the  acoustic  impedance 
over  a  I  percent  range.    These  components 
are  then  assembled  in  a  test  block  to 
provide  interfaces  of  varying  reflectivity 

fluid  (DC200-20CS) .    This  mixture  provided  a 
range  of  acoustic  impedances  (fig.  3).  Suspended 
in  the  silicone  fluid  were  particles  whose  mean 
diameter  was  7  pm.    These  particles  represented 
a  source  of  acoustic  scattering,  thus  an  ultra- 
sound B-mode  scan  of  the  test  block  revealed  the 
interface  between  silastic  blocks  of  various  com- 
position and  scattering  which  occurs  in  the  bulk 
medium  (fig.  4).    Panels  A,  B,  C  and  D  show  the 
test  block  at  four  gain  levels  40,  30,  20  and 
10  dB.    An  adjustment  was  then  madd  to  the  visuali- 
zation system  which  sacrifices  sensitivity  for 
higher  longitudinal  resolution,  better  signal-to- 
noise  ratio  and  dynamic  range  compression,  but  dis- 
plays the  weak  specular  reflectors  at  the  inter- 
faces with  good  "gray  scale"  and  resolution,  but 
without  displaying  the  scatterers.    Panel  E  shows 
the  effect  of  optimizing  the  system  response  for 
visualizing  the  interfaces  within  the  test  block. 
The  response  curve  was  adjusted  to  give  an  output 
of  1  volt  for  an  interface  with  1.0  percent  dif- 
ference in  acoustic  impedance  (minimum  detectable 
brightness  on  the  display).    The  echoes  from  the 
scatterers  then  fell  below  0.1  volt  level.  This 
is  an  example  of  the  way  in  which  the  test  object 
can  be  used  to  standardize  the  equipment  perfor- 
mance so  as  to  optimize  the  display  of  a  particu- 
lar type  of  target,  organ  or  condition. 

Plastisol  Phantoms  -  Phantoms  constructed  of 
plastisol  are  more  suitable  for  basic  studies  on 
the  interaction  of  sound  and  tissue  because  of 
their  inherent  stability  and  the  precision  with 
which  the  characteristic  acoustic  impedance, 
velocity  and  density  can  be  controlled  to  achieve 
precise  acoustic  properties.    Because  of  its  re- 
latively high  attenuation  it  is  not  ideally  suited 
for  construction  of  training  phantoms  where  prop- 
erties and  distances  of  15  to  20  cm  are  required. 

By  using  various  plasticizers  and  by  adjust- 
ing the  resin/plasticizer  ratio,  it  is  possible 
to  modify  the  characteristic  acoustic  impedance 
of  the  system  throughout  the  range  of  soft  tissue 
structures.    The  acoustic  properties  of  a  system 


E 


Fig.  4.    Ultrasonic  scan  (at  various  sensitivities) 
of  test  block  shows  low  intensity  specular 
reflections  at  interfaces  and  scattering 
from  particles  in  the  bulk  material. 

using  epoxy  soya  plasticizer  are  shown  in  figure 
5.    Also  plotted  are  a  number  of  tissue  charac- 
teristics for  reference.    It  will  be  noted  that 
it  is  possible  to  match  the  impedance,  velocity 
or  density  of  most  any  tissue  and  that  the  other 
parameters  will  be  nearly  equivalent.    The  at- 
tenuation of  this  plastisol  system  is  higher  than 
tissue  and  ranges  from  about  2  dB  cm"i  MHz'i  to 
4  dB  cm"i  MHz"i  for  the  materials  shown  in 
figure  6. 

Analysis  of  the  frequency  characteristics  is 
achieved  by  making  measurements  with  a  range  of 
transducers  operating  in  a  pulsed  mode.    A  Fast 
Fourier  Transform  (FFT)  for  displaying  the  fre- 
quency spectrum  of  the  pulse  transmitted  through 
the  specimen  is  compared  with  the  spectrum  ob- 
tained for  a  pulse  transmitted  through  water 
[13].    Figure  7  shows  the  normalized  frequency 
response  for  both  the  reference  and  sample  wave- 
forms.   Amplitude  comparison  between  the  reference 
and  sample  transformed  responses  yields  the  nor- 
malized attenuation  as  a  function  of  frequency, 
as  shown  in  figure  8.    Additional  transducers  may 
be  utilized  to  extend  the  frequency  range.  Phase 
comparisons  between  the  reference  and  sample  re- 
sponses give  the  phase  velocity,  as  plotted  in 
figure  9. 


330 


1.15  - 


1.10 


1.05  - 


1 .00 


0.95 


1 .45 


Fig.  5.    Acoustic  properties  of  a  plastisol  system 
in  which  the  resi n/pl asti ci zer  ratio  is 
adjusted  to  match  the  impedance  of  bio- 
logical materials. 


n.e  5. 

FREOUENC 


S.e  7.3 

MEGftHERTZ 


Fig.  7. 


2  3  4 

Frequency  (MHz) 

Fig.  6.    The  normalized  attenuation  as  a  function 
of  frequency  is  shown  for  two  plastisol 
samples  having  impedances  of  1.52  and  1.63 
X  105  Rayl  r^espectively.    Note  that  the 
former  closely  matches  the  impedance  of 
water  so  that  the  reflection  losses  at  the 
interface  are  minimal.    In  the  second 
specimen,  the  reflection  losses  average 
1.4  dB  per  surface.    Inasmuch  as  most 
normal  soft  tissue  has  a  much  lower  at- 
tenuation rate  than  that  shown  for  these 
two  samples,  materials  of  this  type  would 
be  utilized  to  represent  highly  attenuat- 
ing pathological  states  such  as  those  com- 
monly found  in  breast  tumors. 


The  method  for  evaluating  tissue  simula- 
tors includes  measurement  of  the  fre- 
quency response  characteristics  of  the 
specimen  using  a  computer-generated 
Fourier  transform  of  the  pulse  propagated 
through  the  specimen.    The  S201  sample 
used  here  is  from  the  epoxy  soya  plastisol 
family  having  an  acoustic  impedance  of 
about  1.56  X  10^  Rayl.    The  upper  curve, 
"S",  is  the  reference  response  without 
the  sample.    Below  it  are  four  sample 
responses  for  four  closely  spaced  points 
on  the  sample.    The  signal-to-noise  ratio 
for  this  run  is  such  that  the  data  in  the 
3  to  6  MHz  range  is  valid,  with  the  best 
accuracy  assumed  around  4.2  MHz,  the  fre- 
quency of  maximum  sample  signal  amplitude. 
The  sample  and  reference  responses  may  be 
compared  both  in  magnitude  and  phase. 
Magnitude  comparison  yields  the  sample 
attenuation  (see  fig.  8)  while  phase  com- 
parison gives  the  sample  phase  velocity 
(see  fig.  9) . 


Scatterers  -  The  vast  majority  of  echoes  ob- 
tained by  scanning  a  patient  are  from  scatterers 
rather  than  specular  reflectors.    These  scatter- 
ers produce  echoes  which  are  not  individually  re- 
solved but  add  constructively  when  they  are  in 
phase  to  produce  a  spot  on  the  display.    To  dupli- 
cate this  phenomenon  in  tissue  phantoms,  it  is 
necessary  to  embed  small  scatterers  (like  the 
cells  in  tissue)  whose  dimensions  are  small  com- 
pared to  the  wavelength  of  the  sound.  Spherical 
scatterers  adequately  represent  cells  whose  geome- 
try is  essentially  spherical,  but  elongated  scat- 
terers are  required  to  represent  tissues  in  which 


331 


51 
O 

X 

2: 


2; 
o 


c. 


) 

f 

•> 

*^ 

I— 

, — 4; 

i 

I 

3.0 


3.6  4.2 

FREQUENCY 


4.8  5.4 
MEGAHERTZ 


6.C 


Fig.  8. 


The  attenuation,  normalized  in  the 
customary  manner,  for  the  Fast  Fourier 
Transform  responses  given  in  figure  7, 
is  given  here.    Only  data  in  the  assumed 
valid  data  range  of  3  to  6  MHz  is  pre- 
sented; however,  the  variation  in  attenua- 
tion seen  here  is  still  partly  due  to  the 
low  signal  amplitudes  near  3  and  6  MHz. 
The  solid  bar  at  the  bottom  of  the  plot 
indicates  the  range  of  frequencies  over 
which  a  least  squares  curve  fit  was  per- 
formed on  the  attenuation  (in  dB)  vs_. 
frequency  data.    The  arrow  is  on  the  fre- 
quency of  maximum  amplitude,  as  described 
in  figure  7. 


the  cellular  makeup  is  other  than  spherical,  like 
in  muscle,  for  example. 

Small  plastic  microspheres  are  available  in  a 
variety  of  material  types  which  can  be  embedded 
in  the  acoustic  medium  to  produce  scattering. 
The  particles  come  in  a  variety  of  sizes.  In 
general  the  particle  size  should  approximate  the 
dimensions  of  the  scatterers  of  tissue,  if  the 
frequency  dependence  of  scattering  is  important. 

One  of  the  common  sources  is  the  Dextran  par- 
ticles (Sigma  Chemical  Company,  for  example)  used 
in  Sephadex  columns.    This  type  of  particle  tends 
to  imbibe  water,  thus  the  dimension  in  the  dry 
state  will  be  different  than  the  dimension  in  the 
aqueous  medium.    Particles  can  be  sized  by  putting 
the  powder  through  graded  sieves.    Waag  et  al . 
[14]  have  used  three  grades,  each  of  which  con- 
tains a  wide  range  of  sizes.    There  may  be  occa- 
sions when  the  range  of  sizes  would  be  desirable 
in  simulating  tissue.    It  is  characteristic  of 
tissues  to  have  rather  uniform  cellular  dimensions. 


Grade 

G50M 
G50F 
G50SF 


Size 

86  -  257  ym 
34  -  137  ym 
17  -    68  ym 


The  uniformity  of  unscreened  microspheres  com- 
pared with  screened  microspheres  is  shown  in 
figure  10.    A  uniform  population  of  microspheres 
may  be  used  to  study  the  frequency  dependence  of 
scattering  and  will  simulate  tissues  having 


05 
\ 


o 
_J 

Ld 
>„ 


>- 

>- 

r 

3.0 


Fig.  9. 


3.6  4.2 

FREQUENCY 


4.8  5.4 

MEGAHERTZ 


The  phase  velocity  of  the  sample  examined 
in  figures  7  and  8  is  presented  here. 
This  phase  velocity  is  defined  by  the 
equation 

Cs  -  (l/C^  +  d(fi/df)(l/2Trt) 

where  C^  is  the  velocity  of  the  reference 
medium,  t  is  the  thickness  through  which 
the  acoustic  pulse  travels  in  the  sample, 
and  d^/df  is  the  slope  of  the  phase  angle 
vs.  frequency  data.    This  latter  value  is 
determined  by  the  least  squares  technique 
over  the  range  of  frequencies  indicated 
by  the  solid  bar  at  the  bottom  of  the 
plot.    The  arrow  is  on  the  frequency  of 
maximum  amplitude,  as  described  in  figure 
7.    For  comparison,  the  group  velocity, 
that  is,  the  velocity  of  the  pulse  con- 
taining all  frequencies  as  shown  by  the 
Fourier  transform  of  figure  7,  for  this 
sample  is  1.558  m/ms. 


spherical  cells.    The  range  of  frequencies  for 
the  transition  from  Rayleigh  to  Bragg  scattering 
is  very  broad  if  the  distribution  of  micro- 
spheres as  shown  in  figure  lOA  is  used  in  a  tis- 
sue phantom,  but  is  abrupt  with  scatterers  in 
figure  lOB.    This  point  will  be  further  developed 
later  (see  fig.  14). 

Acrylic  microspheres  are  available  as  a  dental 
repair  material.    These  microspheres  can  also  be 
sifted  through  graded  sieves  to  obtain  uniform 
size  distribution.    In  some  instances  the  more 
intense  scattering  provided  by  the  acrylic  micro- 
spheres is  desirable.    Further,  the  dimensional 
stability  is  very  much  greater,  therefore,  for 
applications  in  which  dimensional  stability  is 
crucial,  the  acrylic  microsphere  should  be  con- 
sidered. 

Another  source  of  microspherical  particles  is 
Teflon  microspheres,  which  are  similar  to  the 
acrylic  microspheres  in  terms  of  scattering, 
strength  and  dimensional  stability.    Both  the 
Teflon  and  acrylic  microspheres  may  need  to  be 
mixed  with  a  wetting  agent,  such  as  alcohol,  to 
promote  suspension.    Since  these  spheres  are 
denser  than  water,  they  tend  to  settle  out  faster 
than  the  Dextran. 


Fig.  10.    The  Sephadex  column  packing  spheres  are 
useful  sources  of  acoustic  scatterers 
and  are  available  in  three  size  ranges; 
however,  each  of  the  size  ranges  has  a 
spectrum  of  scatterer  sizes.    The  photo- 
graph on  the  left  shows  the  distribution 
of  sizes  obtained  for  the  medium  range. 
The  photograph  on  the  right  shows  the 
20  ym  sizes  selected  by  a  screening 
process . 

Hollow  glass  microspheres  show  the  greatest 
difference  in  impedance  compared  with  the  medium 
(fig.  IIA).    The  glass  microspheres  are  lower  in 
density  than  water  and  tend  to  float. 

Between  these  four  spherical  scatterers  with 
their  various  acoustic  properties,  it  is  possible 
lj  to  simulate  many  different  tissue  conditions. 
?  Scatterers  are  added  to  the  medium  in  a  concentra- 
I   tion  of  approximately  1  percent  per  unit  volume. 
'  For  Dextran  this  gives  a  backscattering  charac- 
r  teristic  equivalent  to  a  highly  reflecting  tissue, 
■j  In  many  cases  less  than  1  percent  is  adequate  to 
'•mimic  the  desired  tissue. 

The  angular  dependence  of  scattering  can  reveal 
information  on  the  orientation  of  asymmetrical 
I  scatterers.    It  is  therefore  important  to  include 
ji  such  sources  in  the  test  target  that  are  aligned 
in  a  specific  direction.    Targets  containing  such 


Fig.  11.    Scattering  particles  consist  of  cellulose 
acetate  filaments  or  plastic  spheres. 
The  acetate  fibers  are  5  um  in  diameter 
and  0.1  mm  in  length,  and  the  spheres 
shown  in  the  photograph  are  approximately 
10  pm  in  diameter,  but  other  sizes  can 
readily  be  obtained  by  putting  the  random 
spheres  through  screens  of  the  appropriate 
size  mesh.    Alignment  of  the  fibers  is 
achieved  by  flowing  the  medium  prior  to 
curing.    The  number  of  scatterers  per  unit 
volume,  the  size  of  the  scatterers,  the 
difference  in  acoustic  index  of  refraction 
of  the  scatterers  compared  with  the  em- 
bedding medium,  the  form  factor,  the 
diameter/length  ratio  and  the  degree  of 
order  are  all  factors  which  influence  the 
statistical  distribution  of  scattering. 

scatterers  will  exhibit  angular  dependence  of  scat- 
tering more  like  that  of  muscle. 

Rod-like  scatterers  can  be  obtained  by  using 
flocking,  which  is  a  cellulose  acetate  material 
(see  fig.  UB).    The  flocking  is  an  acetate  fila- 
ment and  is  available  in  various  size  ranges. 
Alignment  of  the  rods  can  be  achieved  by  causing 
the  material  to  flow,  then  gelling  the  substrate 
before  diffusion  reorients  the  rods.    The  rods 
exhibit  an  angular  dependence  of  scattering  when 


333 


they  are  all  aligned  in  the  same  direction. 
Flocking  is  available  in  a  variety  of  size  ranges, 
but  typical  dimensions  are  15  micrometers  by  1 
mi  1 1 imeter. 

Attenuators  -  The  attenuation  of  the  hydrocol- 
loid  materials  is  low  and  comparable  to  water. 
The  addition  of  scatterers  increases  the  attenua- 
tion, but  this  may  still  be  below  the  level  needed 
to  simulate  many  tissues;  therefore,  it  is  neces- 
sary to  add  a  substance  to  further  increase  the 
attenuation  without  increasing  the  scattering  prop- 
erties.   A  substance  to  achieve  this  property  would 
be  any  insoluble  powder  whose  particle  size  is  very 
small  compared  with  the  wavelength  of  the  examining 
beam.    Numerous  powders  will  fulfill  this  require- 
ment.   Among  those  used  is  diatomaceous  earth,  tal- 
cum powder,  powdered  chalk,  graphite,  lamp  black, 
etc .    We  have  used  diatomaceous  earth  for  this  prin- 
cipally because  of  its  availability.    The  powder 
is  mixed  with  the  scattering  spheres  and  hydrocol- 
loid  until  the  desired  attenuation  has  been  achiev- 
ed.   Using  this  technique  the  scattering  and  at- 
tenuation can  be  adjusted  independently  over  some 
range  of  acoustic  characteristics. 

Fabrication  Techniques  -  Organs  can  readily  be 
molded  by  using  natural  organs  or  models  of  natural 
organs.  Any  of  several  molding  techniques  is  suit- 
able, including  gel  molds.  The  positive  replica  of 
the  organ  is  cast  in  the  gel  and  the  mold  is  par- 
tially opened.  The  replica  is  extracted  by  deform- 
ing the  gel  mold. 

The  surface  of  the  mold  may  be  covered  with 
glycerin  or  other  suitable  parting  agent  and  the 
warm  sol  solution  is  poured  into  the  mold.  Sprues 
can  be  provided  to  carry  away  gas  as  the  gas  in 
the  mold  is  displaced  by  the  sol. 

If  the  organ  is  to  represent  a  kidney,  for  in- 
stance, it  would  be  desirable  to  cast  the  organ  in 
several  stages  to  build  up  a  structure  which  simu- 
lates the  internal  architecture  of  the  kidney.  In 
this  case  the  core  is  held  in  place  by  struts  while 
the  cortex  is  b^ing  cast  in  the  mold.  Afterwards, 
the  struts  are  removed  and  the  void  is  filled  with 
sol,  using  a  syringe.    The  completed  kidney  is  then 
encapsulated  in  a  film  to  minimize  evaporation  and 
add  stability  and  convenience  of  handling.  The 
film  can  be  applied  by  dipping  or  by  sealing  the 
material  with  such  as  a  thin  polyethylene  over  the 
surface  of  the  organ  and  heat  sealing  the  edges. 
The  very  thin  polyethylene  is  easily  stretched  over 
the  organ  and  if  the  procedure  is  carefully  done, 
it  is  possible  to  avoid  inclusion  of  air  between 
the  skin  and  the  kidney.    The  thickness  of  the  skin 
will  determine  the  specular  reflection  from  the 
surface  of  the  organ. 

Specular  reflectors  can  also  be  generated  by 
casting  smooth  interfaces  between  materials  having 
different  acoustic  impedances  and  scatterers  are 
introduced  by  mixing  plastic  microspheres  or  rods 
in  the  compound  before  the  heat  treating  (solva- 
tion).   The  plastic  microspheres  should  be  of  a 
size  which  is  very  small  compared  to  the  wave- 
length and  comparable  in  dimensions  to  cells  in 
tissue,  i.e.,  in  the  order  of  10  micrometers  in 
diameter.    The  material  of  the  microsphere  could 
be  chosen  so  that  it  will  not  be  affected  by  the 
embedding  material.    In  order  to  avoid  introducing 
gas  bubbles  into  the  mixture  it  is  necessary  to 
wet  the  microspheres  with  a  small  quantity  of 
liquid  which  is  miscible  with  the  plasticizer  and 
which  does  not  interfere  with  its  properties, 
such  as  alcohol . 


3.    Classification  of  Tissue  Simulators 

As  noted  in  the  Introduction  there  are  three 
major  applications  for  tissue  simulators,  or  phan- 
toms:   1)  a  training  device  to  be  used  in  place  of 
human  subjects;  2)  a  device  for  evaluating  the 
performance  of  ultrasonic  visualization  systems; 
and  3)  a  model  for  studying  the  basic  interactions 
between  sound  and  tissue. 

Each  of  the  three  areas  of  application  imposes 
design  requirements  on  the  construction  of  tissue 
simulators.    The  training  device  should  yield 
realistic  echograms  when  scanned  with  conventional 
diagnostic  equipment  so  that  the  trainee  can  learn 
scanning  techniques  and  the  interpretation  of  echo- 
grams.   The  training  device  should  also  be  fabri- 
cated to  include  both  normal  and  abnormal  anatomy 
(fig.  12)  illustrating  the  diagnostic  criteria. 
The  phantom  should  contain  all  of  the  necessary 
landmarks  and  reference  points  used  by  ultrasono- 
graphers  in  their  diagnostic  protocol.    The  dimen- 
sions of  the  training  device  should  hold  a  1:1 
correspondence  with  the  dimensions  of  the  struc- 
tures represented,  therefore,  the  attenuation  in 
the  tissue  simulator  should  correspond  closely 
with  the  attenuation  experienced  in  the  correspond- 
ing human  anatomy. 


Fig.  12.    Simulation  of  a  cross-sectional  slice  of 
the  human  torso  which  can  be  scanned  with 
an  ultrasonic  transducer.    The  resulting 
echograms  exhibit  a  1:1  correspondence  to 
the  same  section  through  the  human  body. 
The  training  device  would  be  constructed  » 
to  show  both  "normal"  anatomy  and  simulate! 
pathology.  ' 

II 

Tissue  simulators  used  to  check  the  performance  i, 
of  ultrasonic  diagnostic  equipment  should  include  >. 
means  for  examining  the  sensitivity  and  resolution  c 
(both  lateral  and  longitudinal)  of  the  equipment  ; 
over  the  dynamic  range  of  signal  intensities  en-  li 
countered  in  clinical  applications  (fig.  13).  It 
is  desirable  that  these  test  objects  be  stable  ' 
with  respect  to  time  so  that  instabilities  in  the 
diagnostic  equipment  can  be  measured.  Using  the  i 
simulator  as  a  standard,  equipment  adjustments  can  i} 


334 


10 


Fig.  13.    A  test  block  incorporating  low  re- 
flectivity specular  reflectors, 
scatterers  and  wedges  for  resolution 
measurements  are  cast  in  a  plastic 
block  which  can  be  utilized  to  examine 
the  performance  of  visualization 
equipment. 

be  made  to  optimize  the  system  performance  for 
displaying  certain  tissue  characteristics,  such  as 
was  done  in  optimizing  the  display  of  weak  scatter- 
ing interfaces  in  figure  4,  panel  E. 

Tissue  simulators  used  for  basic  studies  will 
vary  widely  in  their  design  as  dictated  by  the  ex- 
perimental conditions  under  study.    We  have  been 
using  simple  disc-shaped  test  objects  of  unifonri 
thickness  with  known  scattering  densities  and  im- 
pedance characteristics  to  examine  the  angular  and 
frequency  dependence  of  scattering  in  tissue.  It 
is  important  to  control  the  size  of  the  scatterer. 
For  scatterers  which  are  very  small  compared  to 
the  wavelength  of  sound,  scattering  is  of  the  Ray- 
leigh  type  and  has  a  fourth  pov/er  cross  section. 
In  the  range  of  Bragg  scattering,  there  is  a  power 
of  two  dependence  of  frequency  on  the  scattering 
cross  section  (fig.  14).    This  plot  is  an  extension 
of  data  presented  by  Freese  et  al .  [15]  and  nicely 
illustrates  the  regions  of  Rayleigh  and  Bragg  scat- 
tering for  oil  droplets  in  gelatin  and  lipid  filled 
cells  in  fish  muscle. 

4.  Conclusion 

Tissue  simulators  are  necessary  to  meet  various 
requirements  with  respect  to  the  application  of 
ultrasound  to  medicine  which  are  not  fulfilled  by 
test  targets  such  as  the  AIUM  100  mm  test  object. 
Recently  there  has  been  an  interest  in  developing 
tissue  equivalent  targets  from  several  sources 
and  this  interest  stems  from  both  a  requirement 
to  provide  more  quantitative  diagnostic  methods 
as  well  as  an  interest  in  reducing  unnecessary  ex- 
posure of  patients  to  ultrasound.    Although  as  yet 
no  toxic  effects  have  been  documented  for  visuali- 
'zation  systems  used  in  the  customary  manner,  it  is 
prudent  to  minimize  unnecessary  exposure  whenever 
possible.    Because  it  is  possible  to  accurately 
simulate  body  structures  for  ultrasonic  scanning, 
construction  of  such  phantoms  would  appear  to  be 
an  important  current  objective. 


.01  - 


.001  - 


.0001 


(coho)muscle 
with  2%  lipid  content 


\im  droplet 


backscatter  coefficient 
Q  for  oil  droplets  in 
gelatin  as  a  function 
of  frequency 


I   I  I  I  I 


Fig.  14. 


1  10  100 

Frequency  (MHz) 

The  plot  of  the  backscatterers '  cross 
section  vs^.  frequency  for  oil  droplets 
in  gelatin  and  for  lipid  droplets  in 
muscle  display  both  Rayleigh  and  Bragg 
scattering  phenomena.    The  Rayleigh 
scattering  has  a  fourth  power  dependence, 
whereas  the  Bragg  scattering  has  a 
second  power  dependence  on  frequency. 


Acknowl edgments 

This  work  was  supported  in  part  by  NSF  Grant 
APR75-15908,  a  Smith-Kline  Fellowship,  and  by  The 
Indianapolis  Center  for  Advanced  Research.  The 
computer  processing  technique  was  developed  by 
Francis  J.  Fry  and  Narendra  T.  Sanghvi. 

References 

[1]    Stanford,  E.  C.  C,  J.  Soc.  Chem.  Ind.  5, 
218  (1886).  ~ 

[2]    Nelson,  W.  L.  and  Cretcher,  L.  H.,  J.  Amer. 
Chem.  Soc.  51_,  1914  (1929). 

[3]    Atkins,  E.  D.  T.,  Mackie,  W.,  and  Smolko, 
E.  E.,  Nature  225,  626  (1970). 

[4]    Atkins,  E.  D.  T.,  Mackie,  W. ,  Parker,  K.  D., 
and  Smolko,  E.  E.,  Polymer  Lett.  9,  311 
(1971). 

[5]    Rees,  D.  A.  and  Samuel,  J.  W.  B.,  J.  Chem. 
Soc.  C,  2295  (1967). 

[6]    Larsen,  B.,  Painter,  T. ,  Haug,  A.,  and 
Smidsr^d,  0.,  Acta  Chem.  Scand.  23,  355 
(1969). 

'     Painter,  T. ,  Smidsreid,  0.,  Larsen,  B.,  and 
Haug,  A.,  Acta  Chem.  Scand.  22,  1637  (1968). 

[8]    McDowell,  R.  H. ,  J.  Soc.  Chem.  Ind.  (London), 
Monograph  No.  24,  19  (1966). 


335 


[9]    Hirst,  E.  L.,  Percival,  E.,  and  Wold,  J.  K., 
Chem.  Ind.  (London),  257  (1963). 

[10]    Smidsrjid,  0.  and  Haug,  A.,  Acta  Chem.  Scand. 
19,  329  (1965). 

[11]  Skinner,  E.  W.  and  Phillips,  R.  W.,  eds.. 
The  Science  of  Dental  Materials,  pp.  101- 
135  (W.  B.  Saunders  Co. ,  Philadelphia,  1967). 

[12]    Eggleton,  R.  C,  Fabrication  of  Ultrasonic 
Test  Specimen  for  Evaluating  the  Performance 
of  Echo-Ranging  Equipment,  submitted  to  AIUM 
Standards  Committee  (1970). 

[13]    Franklin,  T.  D.,  Jr.,  Sanghvi,  N.  T.,  Fry, 
F.  J.,  Egenes,  K.  M.,  and  Weyman,  A.  E., 
Ultrasonic  tissue  characterization  studies 
of  ischemic  and  infarcted  myocardium, 
presented  at  2nd  International  Symposium  on 
Ultrasonic  Tissue  Characterization,  Session 
5;  June  13-15,  1977,  Gaithersburg,  Maryland. 

[14]    Waag,  R.  C.  et  al . ,  personal  communication 
(advanced  copy  of  manuscript). 

[15]    Freese,  M.  L.  and  Hamid,  M.  A.  K.,  Lipid 
content  determination  in  whole  fish  using 
ultrasonic  pulse  backscatter,  in  1974  Ultra- 
sonics Symposium  Proceedings,  pp.  69-76, 
IEEE  Cat.  No.  74  CH0896-1SU. 


336 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed.,  National  Bureau 
of  Standards,  Spec.  Publ.   525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


TISSUE  EQUIVALENT  TEST  OBJECTS  FOR  COMPARISON  OF  ULTRASOUND 
TRANSMISSION  TOMOGRAPHY  BY  RECONSTRUCTION  WITH  PULSE 
ECHO  ULTRASOUND  IMAGING 


Paul  L.  Carson  and  Leonard  Shabason 

Department  of  Radiology 
University  of  Colorado  Medical  Center 
Denver,  Colorado    80020,  U.S.A. 

Donald  E.  Dick 

Department  of  Physical  Medicine 
University  of  Colorado  Medical  Center 
Denver,  Colorado    80020,  U.S.A. 

and 

Wi  1 1  iam  dayman 

Alderson  Research  Laboratories,  Inc. 
Stamford,  Connecticut    06903,  U.S.A. 


Tissue  equivalent  materials  have  been  investigated  for  evaluation  and  comparison  of 
pulse  echo  ultrasound  imaging  and  ultrasound  transmission  tomography  by  reconstruction 
(UTTR).    Investigations  have  centered  primarily  on  various  urethane  polymers  and  3M 
Reston  Brand  Flotation  Pad  material.    Attenuation  coefficients  of  the  urethane  polymers 
still  are  somewhat  too  high,  and  thus  initial  test  objects  or  "phantoms"  have  been  con- 
structed from  the  flotation  pad  material.    One  phantom  chosen  to  simulate  several  char- 
acteristics of  human  breast  tissue  consists  of  an  annulus  of  unaltered  flotation  pad 
material  surrounding  a  center  region  in  which  scattering  polystyrene  microspheres  are 
embedded.    Contrasting  material  such  as  polyethylene  rods  can  be  inserted  between  the 
inner  and  outer  areas  of  the  phantom.    The  UTTR  technique  clearly  delineates  poly- 
ethelene  rods  in  the  phantom  as  small  as  .6  mm  in  diameter.    The  scattering  at  2.2  to 
3.5  MHz  nominal  frequency  exhibits  a  pulse  echo  appearance,  similar  to  that  of  liver 
tissue,  and  causes  only  approximately  a  5  percent  increase  in  the  attenuation  coeffi- 
cient of  the  pure  flotation  pad  material . 

Key  words:    Computed  tomography-ultrasonic;  diagnosis-ultrasonic;  tissue  equivalent 
test  objects  and  phantoms;  ultrasound  imaging;  ultrasonic  tissue  char- 
acterization. 


1.  Introduction 

Among  the  various  properties  of  tissue  that 
can  be  detected  with  ultrasound,  attenuation  and 
velocity  have  attracted  a  great  deal  of  atten- 
tion.   This  interest  has  been  intensified  by 
evidence  that  malignant  and  benign  breast  tumors 
can  be  distinguished  from  each  other  and  from 
normal  tissue  by  significant  differences  in  at- 
tenuation of  ultrasound  by  these  different  tis- 
sues [1]^.    The  goal  of  the  main  project  lead- 
ing to  this  study  has  been  to  develop  a  system 
for  the  quantitative  estimation  of  the  spatial 


'Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


distribution  of  attenuation  and  velocity  co- 
efficients in  living  tissue.    This  instrumenta- 
tion development  has  been  accompanied  by  the 
development  of  various  tissue-equivalent  test 
objects  or  phantoms  to  help  ensure  accurate  and 
rel iable  results. 

Simultaneous  measurement  of  velocity  and  at- 
tenuation can  be  performed  with  direct  ap- 
proaches to  ultrasound  computerized  tomography 
(CT),  although  many  questions  exist  as  to  the 
accuracy  of  these  measurements.    To  study  the 
potential  of  quantitative  and  visual  diagnosis 
with  the  technique,  a  system  for  performing 
ultrasound  transmission  tomography  by  recon- 
struction (UTTR)  has  been  designed  and  con- 
structed [2].    The  attenuation  image  shown 
here  was  obtained  with  this  new  instrumentation, 


337 


using  principles  delineated  in  reference  [3]. 
Basically,  opposed  transmitting  and  receiving 
transducers  are  scanned  on  either  side  of  the 
imaged  object  in  the  translate-rotate  motion 
of  early  x-ray  CT  scanners.    The  amplitude  of 
the  rectified  and  integrated  pulse  waveform 
was  linearized  and  used  as  the  input  variable 
in  a  convolution  reconstruction  program.  The 
complete  UTTR  system  soon  will  include  the 
capability  for  simultaneous  generation  of 
pulse-echo  images  as  well  as  CT  scans  of  veloc- 
ity, and  attenuation.    The  system  has  enough 
flexibility  to  serve  as  a  general  purpose  re- 
search instrument,  but  the  initial  investiga- 
tions will  focus  on  breast  imaging.    This  em- 
phasis was  chosen  because  of  the  high  incidence 
of  breast  cancer  in  women  coupled  with  the  need 
for  more  reliable,  non-invasive  techniques  to 
detect  and  characterize  breast  lesions.  Another 
factor  is  the  existence  of  preliminary  encourag- 
ing results  obtained  by  others  with  ultrasound 
velocity  CT  images  [4]. 

To  provide  a  controlled  comparison  of  the 
various  imaging  approaches,  tissue  equivalent 
objects  have  been  developed.    This  development 
has  followed  three  approaches  which  have  been 
pursued  concurrently: 

1.  Development  and  identification  of 
stable  plastics  which  are  equivalent 

to  tissue  in  terms  of  physical  density, 
sound  propagation  velocity,  and  atten- 
uation as  a  function  of  frequency. 

2.  Simulation  of  pulse  echo  scattering 
using  small  reflectors  embedded  in 
available  plastics. 

3.  Molding  the  available  tissue  equivalent 
plastics  and  scattering  materials  into 
phantoms  to  allow  evaluation  and  com- 
parison of  alternative  imaging  approaches. 

2.  Developments 

To  date,  most  of  the  investigation  on  tissue 
equivalent  plastic  has  been  directed  to  various 
urethane  polymers  such  as  TDI  polyether  urethane 
prepolymer  with  high  molecular  weight  polyol 
chain-extenders,  and  to  the  3M  Company  material 


used  in  Reston  Flotation  Pads.    The  urethane 
polymers  can  be  made  with  densities  and  ultra- 
sound propagation  velocities  in  the  soft  tissue 
range.    Shown  in  figure  la  is  an  example  of  at- 
tenuation measurements  on  one  of  these  polymers, 
which  has  a  density  and  ultrasound  propagation 
velocity  at  27  ^C  of  1.1  x  10^  kg/m3  and  1544 
m/s,  respectively.    Work  is  progressing  on  reduc- 
ing the  ultrasound  attenuation  coefficient  in 
these  materials  to  that  of  various  soft  tissues. 
Advantages  of  this  class  of  polymer  are  that  the 
materials  are  rigid  enough  to  support  test  ob- 
jects in  known  locations,  properties  can  be 
varied,  and  the  materials  are  stable  in  time. 
The  3M  Flotation  Pad  material  exhibits  ultrasound 
interaction  characteristics  similar  to  that  of 
fat  except  for  a  somewhat  higher  attenuation  co- 
efficient as  shown  in  figure  lb.    The  amplitude 
attenuation  coefficient  in  SI  units  is  related 
to  the  attenuation  coefficient  in  dB/cm  by 
)j(m-M  =  11.513  •  a(dB/cm).    Also,  p/f  (s/m)  = 
1.1513  X  10-- 5  •  a/f  (dB/cm-MHz).    The  speed  of 
ultrasound  propagation  in  the  flotation  pad  ma- 
terial at  27  "^C  is  1461  m/s  and  its  density  is 
9  X  10^  kg/m^.    Figure  1  was  obtained  using 
broadband  transmitters  with  working  frequencies 
of  2.2  and  4  MHz  with  diameters  of  19  and  13  mm, 
respectively.    A  0.25  mm  diameter  miniature 
hydrophone  was  employed  as  a  receiver  in  the  far 
field  of  the  transmitter,  and  attenuation  of 
gated  sine  waves  was  measured  by  interposing  and 
removing  samples  of  the  materials  under  study. 

Polystyrene  microspheres  in  the  range  of  15 
to  300  )im  in  diameter  have  been  embedded  in 
various  plastics,  and  their  pulse  echo  imaging 
characteristics  have  been  studied  in  densities 
of  0.2  to  2.0  spheres  per  um^.    At  these  den- 
sities, 80  ym  diameter  polystyrene  microspheres 
scatter  more  weakly  than  normal  liver,  as  de- 
termined with  commercially  available  pulse  echo 
imaging  systems  at  2.2  -  5  MHz.  Consequently, 
polystyrene  microspheres  of  175  ym  diameter,  in 
densities  of  approximately  0.3  spheres  per  mm^, 
have  been  chosen  for  initial  simulation  of  soft 
tissue  scattering. 

The  phantom  designed  for  comparisons  of  the 
various  ultrasound  imaging  techniques  is  dia- 


f(MHz) 

Fig.  1.    Semilogarithmic  plots  of  attenuation  coefficient  as  a  function  of  frequency  at  two  temperatures 
(a)  urethane  polymer;  (b)  3M  flotation  pad  material. 


338 


Fig.  2.    Top  view  and  side  view  of  the  Tissue 
Equivalent  Test  Object.    The  central 
cylinder  contains  isotropic  scatterers. 
Rods  and  tubes  may  be  inserted  between 
the  central  cylinder  and  the  outer  an- 
nulus.    The  three  nylon  bolts  shown  in 
the  drawing  have  been  replaced  by  0.13 
mm  diameter  nylon  monofilament. 

grammed  in  figure  2.    An  annulus  of  10  cm  outside 
diameter  consisting  of  unaltered  flotation  pad 
material  surrounds  a  4  cm  diameter  of  cylinder  of 
the  flotation  pad  material  containing  the  scatter- 
ing microspheres.    This  5  cm  thick  cylinder  is 
sandwiched  between  two  acrylic  plates  for  geometri- 
cal stability.    Holes  in  the  acrylic  plate  located 
conveniently  at  the  interface  between  the  outer 


annulus  and  the  central  cylinder  allow  rods  and 
liquid-filled  cylinders  of  various  attenuating 
materials  to  be  inserted  between  the  scattering  and 
nonscattering  materials. 

3.    Examples  of  Use 

Pulse  echo  images  of  the  phantom  obtained  with 
a  commercial  ultrasound  scanner  are  shown  in 
figure  3.    Polyethylene  rods  of  1.9,  2.7  and  3.3 
mm  diameter  are  displayed  from  left  to  right  at 
the  edge  of  the  inner  cylinder.    In  this  image, 
the  polyethylene  rods  are  in  the  focal  zone  of 
the  3.5  MHz  transducer  and  the  shadowing  of  the 
scattering  material  by  the  rods  is  visualized 
easily.    Figure  4  is  an  UTTR  image  of  attenuation 
in  the  same  phantom  with  rods  of  3.3,  2.7,  1.9 
and  0.6  mm  diameter.    As  may  be  seen,  each  of  the 
four  rods  was  delineated  clearly.    Phase  cancella- 
tion at  the  receiver  due  to  the  increased  speed 
of  sound  in  the  rods  may  have  contributed  to 
their  detectabi 1 ity  in  both  the  attenuation  image 
and  the  pulse  echo  shadowing  of  discrete  scat- 
terers.    The  same  effect  should  contribute  to 
the  detection  of  highly  attenuating  lesions  in 
the  body. 

Attenuation  coefficients  for  the  flotation  pad 
material  obtained  in  this  reconstruction  are 
a  =  2.88  ±  0.01  dB/cm  or  p  =  33.2  ±  0.2  m-i.  The 
quoted  errors  are  the  standard  error  of  the  mean 
of  several  hundred  pixels  in  the  reconstructed 
image.    These  attenuation  coefficients  correspond 
to  a/f  =  0.82  dB/cm-MHz  at  the  3.5  MHz  working 
frequency  employed  for  the  reconstruction.  If 
the  frequency  averaged  attenuation  coefficient 
of  a/f  =  0.9  dB/cm-MHz  at  21  °C  is  accepted  from 
figure  lb  as  the  true  value  for  the  flotation  pad 
material,  then  the  effective  frequency  of  the 


Fig.  3.    Pulse  echo  image  of  the  tissue  equiva- 
lent phantom  using  a  conventional  scan- 
ner with  a  3.5  MHz  transducer  focused 
at  the  top  side  of  the  central  cylinder 
containing  scattering  particles. 
Shadowing  of  the  scatterers  by  3.3,  2.7, 
and  1.9  mm  polyethylene  rods  is  apparent. 


Fig.  4.    Attenuation  image  of  the  phantom  by 

ultrasound  transmission  tomography  by 
reconstruction.    The  polyethylene  rods 
visible  in  the  image  range  in  diameter 
from  0.6  to  3.3  mm. 


339 


broadband  ultrasound  pulses  used  to  create  the 
UTTR  image  may  be  taken  to  be  3.2  MHz.    By  com- 
parison, the  mean  attenuation  coefficient  of  the 
flotation  pad  material  containing  scatterers  is 
3.02  ±  0.02  dB/cm.    This  measured  value  in  the 
central  scattering  region  may  be  slightly  low  due 
to  beam  hardening  effects.    Nevertheless,  the 
measured  difference  in  the  attenuation  coefficient 
of  the  material  with  and  without  scatterers  is 
only  5  percent,  and  the  real  difference  probably 
is  not  much  greater.    The  greater  attenuation  in 
the  scattering  material  is  visualized  easily  in 
UTTR  images  displayed  at  high  contrast. 

The  maximum  reconstructed  attenuation  coeffi- 
cients at  the  centers  of  the  3.3,  2.7,  i.9  and 
0.6  mm  diameter  polyethylene  rods  are  20,  12,  12 
and  6  dB/cm,  respectively.    Clearly,  these  ap- 
parent attenuation  coefficients  for  the  smaller 
rods  are  reduced  by  resolution  effects.  Bulk 
measurements  of  attenuation  coefficients  of  the 
polyethylene  rods  were  not  obtained,  because  the 
rods  were  made  by  heating  and  stretching  larger, 
low  density  polyethylene  rods.    This  process 
probably  would  affect  the  density  of  the  poly- 
ethylene, and  consequently  the  ultrasound  attenua- 
tion coefficient. 

Pulse  echo  imaging  of  large  samples  of  the 
flotation  pad  material  containing  polystyrene 
microspheres  has  indicated  that  the  flotation 
pad  material  at  21  °C  attenuates  ultrasound  some- 
what more  rapidly  than  normal  liver  tissue  in 
vivo.    It  was  assumed,  therefore,  that  the  flota- 
tion pad  phantom  would  be  a  rigorous  test  of  UTTR 
and  pulse  echo  systems  for  breast  imaging.  Ini- 
tial measurements  on  the  female  breast  in  vivo, 
indicate  that  there  are  relatively  large  quanti- 
ties of  tissues  in  many  young,  female  breasts 
which  are  much  more  attenuating  than  anticipated. 
These  materials,  presumably  dense,  fibrous  tis- 
sues, often  may  be  aligned  parallel  to  the  skin 
surface  and  thus  provide  a  long,  attenuating 
path  for  transmission  imaging  in  coronal  sections 
of  the  breast.    Attenuation  coefficients  averag- 
ing 15  dB/cm  in  these  highly  attenuating  tissues 
have  been  measured  with  the  3.5  MHz  pulses.  It 
should  be  noted,  however,  that  much  of  the  at- 
tenuation obtained  in  the  UTTR  system  is  due  to 
phase  cancellation  and  refraction  at  the  inter- 
face between  tissues  or  phantom  materials  with 
relatively  low  and  relatively  high  speeds  of 
ultrasound  propagation. 


The  breast  phantom  developed  to  date  will  be 
useful  for  comparing  UTTR  and  pulse  echo  imaging 
under  the  best  conditions,  but  much  more  complex 
phantoms  will  be  required  to  simulate  many  im- 
portant features  of  normal  and  abnormal  breasts. 

Acknowl edgments 

This  investigation  was  supported  in  part  by 
grants  APR  76-05944  from  the  National  Science 
Foundation  and  5T32  CA  9073  awarded  by  the 
National  Cancer  Institute,  Department  of  Health, 
Education,  and  Welfare. 

We  gratefully  acknowledge  the  contributions 
of  the  University  of  Colorado  Medical  Center 
personnel:    Will  Carter,  Tom  Oughton,  Mary  Dick, 
Elliott  Bayly,  Gary  Thieme,  and  William  R. 
Hendee,  as  well  as  the  contributions  of  Paul  E. 
Hansen,  of  the  3M  Company,  who  provided  sample 
materials  and  useful  information. 

References 

[1]    Calderon,  C,  Vilkomerson,  D.,  Mezrich,  R., 
Etzold,  K.  F.,  Kingsley,  B.,  and  Haskin,  M.  , 
Differences  in  the  attenuation  of  ultra- 
sound by  normal,  benign  and  malignant 
breast  tissue,  J.  Clin.  Ultras.  4,  249-254 
(1976). 

[2]    Dick,  D.  E.,  Bay,  H.  P.,  and  Carson,  P.  L., 
Hardware  Design  of  an  Ultrasound  CT  Scanner, 
Proceedings ,  Rocky  Mountain  Biomedical  En- 
gineering Symposium,  published  April  1977 
by  the  Instrument  Society  of  America. 

[3]    Carson,  P.  L.,  Oughton,  T.  V.,  Hendee,  W.  R., 
and  Ahuja,  A.  S.,  Imaging  soft  tissue 
through  bone  with  ultrasound  transmission 
tomography  by  reconstruction.  Medical  Physics 
4,  302-309  (1977). 

[4]    Greenleaf,  J.  F.,  Johnson,  S.  A.,  Samayoa , 
W.  P.,  Duck,  F.  A.,  and  Wood,  E.  H.  ,  Alge- 
braic Reconstruction  of  Spatial  Distributions 
of  Acoustic  Velocities  in  Tissue  from  their 
Time  of  Flight  Profiles,  in  Acoustical 
Holography,  N.  Booth,  ed..  Vol.  6,  pp.  71-90 
(Plenum  Press,  New  York,  1975). 


340 


APPENDIX 


I 


341 


Reprinted  from  Ultrasonic  Tissue  Characterization  II,  M.  Linzer ,  ed..  National  Bureau 
of  Standards,  Spec.  Publ.  525   (U.S.  Government  Printing  Office,  Washington,  D.C.,  1979). 


APPENDIX 


DATA  OF  THE  VELOCITY  AND  ATTENUATION  OF  ULTRASOUND  IN 
MAMMALIAN  TISSUES  -  A  SURVEY 


R.  J.  Parry  and  R.  C.  Chivers 

Physics  Department 
University  of  Surrey 
Guildford,  England 


A  compilation  of  the  reported  values  of  velocity  and  attenuation  of  ultrasound  in 
mammalian  tissues  is  presented  to  give  a  clear  picture  of  the  state  of  knowledge  and 
enable  it  to  be  assessed.    In  the  twenty  years  since  the  last  such  compilation,  the 
progress  in  estimating  the  relative  contributions  of  the  animal  species,  tissue  con- 
dition, temperature  and  frequency  at  which  measurements  were  made  and  the  method  of 
measurement  to  the  observed  variation  has  been  small.    It  is  hoped  that  this  compila- 
tion will  both  be  of  practical  use  and  also  encourage  the  establishment  of  a  stronger 
body  of  fundamental  information  for  the  application  of  ultrasound  in  medicine  and 
biology. 

Keywords:    Attenuation;  mammalian  tissues;  ultrasound;  velocity. 


1.  Introduction 

With  the  increasingly  recognized  importance  of 
fundamental  work  on  the  interaction  of  ultrasound 
with  tissue  there  is  a  need  for  ready  access  to 
currently  available  data  of  the  acoustic  parameters 
of  mammalian  tissues.    There  are  summaries  of  ve- 
locity and  attenuation  data  [1-3]^  but  all  lack 
some  of  the  information  needed  to  make  a  critical 
assessment  of  the  present  state  of  knowledge.  In 
view  of  this  we  have  attempted  here  to  produce  a 
survey  which  is  both  accurate  -  not  always  a  fea- 
ture of  previous  surveys  -  and  useful  in  providing 
such  information. 

The  grouping  of  the  data  is  primarily  by  type 
of  tissue  and  mammalian  species  and  subsequently 
by  tissue  condition  and  temperature.    Tissue  condi- 
tion in  vitro  is  categorized  as  fresh,  fixed  or 
refrigerated  (though  for  comparisons  to  be  proper- 
ly valid  more  precise  details  of  the  history  of  the 
tissues  must  be  known). 

It  is  assumed  that  the  measurements  were  made  at 
sufficiently  low  intensity  such  that  non-linear  ef- 
fects were  not  significant.    However,  only  Dussik 
et  a1 .  [4]  and  Fry  and  Fry  [5]  report  values  for 
ultrasonic  intensity.    This  may  arise  from  the 
present  state  of  the  literature  on  the  measurement 
of  intensity  [6]. 
,        The  quoted  attention  coefficient,  a,  is  defined 
by  A  =  Aoe""^,  where  A  is  the  amplitude  of  a  plane 
i,  wave  of  initial  amplitude  Ag  after  travelling  a 
;'    distance  x  in  the  tissue.    Assuming  a  realistic 
ll    plane  wave,  or  that  any  geometrical  effect  on  the 
I    amplitude  has  been  removed,  this  represents  the 


^Figures  in  brackets  indicate  literature 
references  at  the  end  of  this  paper. 


sum  of  the  energy  absorbed  within  the  tissue  and 
the  energy  scattered  by  inhomogeneities  in  the 
tissue  [7].    It  must  also  be  assumed  that  the  in- 
tensities used  were  low  enough  for  there  to  be  no 
irreversible  changes  caused  in  the  tissue  and  thus 
that  a  is  not  dependent  on  the  intensity.    To  con- 
vert the  values  quoted  to  decibels  they  should  be 
multiplied  by  8.686. 

The  tables  include  the  number  of  measurements, 
N,  and  some  indication  of  the  spread  of  values 
about  the  quoted  mean:    where  possible  the  stand- 
ard deviation  (SO)  or  standard  error  (SE)  of  the 
set  of  measurements,  other'wise  the  range  of  values 
or  some  unspecified  "variation". 

As  a  guide  to  what  may  be  expected  of  the  mea- 
surements reported,  it  was  thought  useful  to  in- 
dicate the  measurement  technique  used.    For  the 
velocity  measurements  these  have  been  divided  in- 
to pulse  transmission  or  reflection  (pulse),  in- 
terferometric  and  sing-around  techniques  and  for 
attenuation  into  broadband,  using  pulses  contain- 
ing only  a  few  cycles  of  vibration,  or  narrow  band, 
with  20  or  more  cycles.    Results  obtained  with 
broadband  pulses  [10-12]  are  usually  quoted  as 
being  at  one  particular  frequency  without  any  de- 
tail of  how  this  frequency  was  selected.    For  pref- 
erence the  frequency  specified  should  be  that  of 
the  maximum  amplitude  in  the  frequency  spectrum  of 
the  pulse.    Recently  the  whole  frequency  spectrum 
of  a  broadband  pulse  has  been  used  to  measure  at- 
tenuation [7,13]  and  there  is  the  possibility  of 
using  swept  frequency  techniques  [60].  Further 
consideration  o1^  measurement  methods  is  out  of 
place  in  the  present  context  and  the  reader  is 
referred  to  the  review  article  by  McSkimin  [8]  and 
the  work  of  Wehr  [9,14]  for  discussion  of  the 
various  techniques  and  the  relative  errors  in- 
volved. 


343 


The  information  collected  in  the  tables  below 
is  that  relating  to  whole  tissues  only.    Thus  much 
data  on  blood  is  excluded  as  it  is  concerned  with 
the  constituents  rather  than  whole  blood  (see  Hus- 
sey  [3]  for  a  summary),  and  in  other  reports  the 
"whole  blood"  is  a  dilution  with  plasma  of  centri- 
fuged  red  blood  cells  [57,59].    Nor  was  it  felt  ap- 
propriate to  include  all  the  data  from  isolated 
references  on  the  variation  of  parameters  with 
other  features  of  the  tissues,  for  example  the 
variation  of  velocity  in  fat  with  water  content 
[15,16]  and  the  work  on  liver  homogenates  [17,18]. 
For  such  information  the  reader  is  referred  to  the 
original  literature. 

Summary 

The  result  of  a  survey  such  as  this  should  per- 
haps be  some  set  of  values  of  parameters  or  curves 
that  may  be  taken  to  be  characteristic  of  the  vari- 
ous tissues  considered.    However,  considering  the 
data  presented  here,  all  that  may  be  concluded  is 
that  at  the  present  time  it  is  virtually  impossible 
to  separate  any  variation  between  different  types 
of  tissue  from  those  variations  due  to  the  many 
other  factors  involved,  such  as  the  measurement 
technique,  temperature,  condition  and  treatment  of 
the  tissue,  the  species  from  which  the  tissue  came 
and  its  pathological  condition.    To  produce  any 
"characteristic"  values  many  assumptions  would  have 
to  be  made,  the  validity  of  which  cannot  be  assessed 
from  the  information  available.    Thus  all  that  the 
present  authors  have  set  out  to  do  is  to  present 
the  data  with  the  relevant  information,  where  pos- 
sible, in  order  that  each  worker  may  draw  his  own 
conclusions  about  average  values  and  their'  limita- 
tions. 

The  table  A,  for  velocity  data  and  B,  for  at- 
tenuation data  contain  this  information;  blank  (--) 
indicates  that  the  appropriate  information  is  not 
stated  in  the  original  paper.    To  examine  the  fre- 


quency and  temperature  dependence  of  velocity,  the 
(a  priori  unwarranted)  assumption  is  made  that  the 
results  for  different  species  and  conditions  may 
be  directly  compared  and  the  measurements  for  two 
temperature  ranges  are  then  represented  in  figures 
1  and  2.    The  ranges  chosen  were  23  +  3  °C  and 
36  ±  1  °C,  the  former  being  rather  wide  in  order 
to  encompass  the  large  body  of  measurements  made  at 
20  °C  and  26  °C.    It  is  left  to  the  reader  to  opine 
about  the  existence  of  velocity  dispersion  or  to 
assess  temperature  dependence  of  velocity  from 
significant  differences  between  figures  1  and  2  - 
a  task  made  difficult  by  the  lack  of  data  at  36  ± 
1  °C  (although  some  data  in  the  tables  does  not  ap- 
pear in  these  figures  as  it  was  averaged  over  a  wide 
frequency  range  [10]).    The  remaining  figures  (3-9) 
show  plots  of  a/f  against  f  for  various  tissues  - 
no  account  being  taken  of  species,  condition  or 
temperature.    This  presentation  was  chosen  as  it 
readily  permits  assessment  of  the  oft-quoted  linear 
relation  between  a  and  f.    Comments  on  the  tempera- 
ture dependence  of  a  are  limited  by  the  consistent 
disinclination  of  authors  to  specify  this  parameter 
when  reporting  their  results  and  their  tendency  to 
average  over  wide  temperature  ranges. 

In  conclusion,  it  is  only  possible  to  echo  Gold- 
man and  Hueter's  statement  of  20  years  ago,  "No 
critical  discussion  is  indicated  at  this  time,  but 
it  is  anticipated  that  the  accumulation  of  further 
data  on  the  basis  of  more  precise  and  extensive 
measurements  should  permit  important  generaliza- 
tions on  the  acoustic  characteristics  of  living 
matter."  [1],  and  in  doing  so  express  the  hope  that 
future  reports  in  the  literature  will  include  suf- 
ficient information  about  tissue  condition  and  mea- 
surement technique  to  enable  valid  conclusions  about 
their  contribution  to  the  spread  of  parameter  values 
to  be  assessed  and  thus  permit  the  "important 
generalizations  on  the  acoustic  characteristics  of 
1 iving  matter. " 


Table.    Compilation  of  (A)  velocity  and  (B)  attenuation  of  various 
tissues  under  various  conditions. 


Species 


Condi  ti  on 


Velocity  of  ultrasound  in  various  tissues  under  various  conditions 

N  Technique 


Temp . 
(°C) 


f 

(MHz) 


(m/s) 


Variation 
(m/s) 


Precision  Source 

it) 


Normal  Blood 


man 

heparini  sed 

22 

4 

2 

1565 

125 

12 

sing-around 

1 

19 

22. 

6 

1570 

24 

10 

23 

2 

1549 

6 

10.8 

10 

24 

2 

1556 

4 

4.7 

11 

Normal  Fat 

man,  orbit 

fresh 

20 

6  -* 

14 

1582 

SD  =  20.4 

65 

pulse 

1 

5 

10 

37 

1462 

SD  =  23.7 

16 

man,  breast 

fresh/dry 

25 

2 

0 

1470 

pul  se 

0 

2 

20 

man,  breast 

refrigerated 

24 

1 

8 

1465 

SE  =  2 

interferometric 

0 

5 

16 

man 

24 

1 

8 

1476 

SE  =  2 

2752 

interferometric 

0 

5 

16 

pig 

fresh 

24 

1 

8 

1444 

SE  =  2 

58 

interferometric 

0 

5 

16 

pig 

fresh 

37 

.5^7 

1440 

21 

cow 

fresh 

24 

1 

8 

1465 

SE  =  3 

21 

i  nterferometri  c 

0 

5 

16 

horse 

fresh 

24 

1 

8 

1443 

SE  =  4 

56 

i  nterferometri  c 

0 

5 

16 

344 


Species        Condition       Temp.         f  c       Variation        N  Technique        Precision  Source 

{°C)       (MHz)      (m/s)       (m/s)  (%) 


Normal  Liver 


man 

— 

24 

1.8 

1585 

SE  =  2 

-- 

i  nterferometri  c 

0. 

5 

16 

pig 

fresh 

24 

1.8 

1587 

SE  =  3 

-- 

i  nterferometri c 

0. 

5 

16 

pig 

fresh 

25 

2.5 

1553 

SD  =  15.5 

7 

sing-around 

0 

5 

22 

cow 

fresh 

24 

1.8 

1590 

SE  =  2 

-- 

i nterferometri c 

0 

5 

16 

cow 

fresh 

24 

1.8 

1578 

SE  =  3 

i  nterferometri  c 

0 

5 

16 

horse 

fresh 

1  Q 
i  .  O 

iooU 

i  nter f erome  trie 

u 

C 

ID 

dog 

fresh 

26 

4 

&  12 

1581 

SD  =  16.8 

8 

i  nterferometri  c 

0 

2 

23 

rabbi  t 

fresh 

24 

1.8 

1599 

SE  =  1 

i nterferometri c 

0 

5 

16 

rabbi  t 

fresh 

26 

4 

&  12 

1575 

SD  =  9.4 

13 

i  nterferometri  c 

0 

2 

23 

guinea  pig 

fresh 

24 

1.8 

1575 

SE  =  3 

i  nterferometri  c 

0 

5 

16 

Abnormal  Liver 

man,  healthy 

refrigerated 

24 

1.8 

1570 

SE  =  4 

18 

interferometric 

0. 

5 

16 

rabbit,  bled 

f  res  h 

9/1 

1 .  o 

1607 

SE  =  2 

1 n eerie  rome ir i c 

u 

D 

1  f\ 

guinea  pig, 
bled 

fresh 

1  o 

1 .8 

1589 
Norma' 

SE  =  2 
Kidney 

interferometri  c 

0. 

5 

16 

pig 

fresh 

24 

1.8 

1560 

SE  =  4 

— 

interferometric 

0 

5 

16 

pig 

fresh 

25 

2.5 

1558 

SD  =  15.6 

5 

si  ng-around 

0 

5 

22 

cow 

fresh 

1.8 

1568 

SE  =  3 

i  nterferometri  c 

U 

c 
3 

i  D 

cow 

fresh 

24 

1.8 

1572 

SE  =  3 

i  nterferometri  c 

0 

5 

16 

horse 

fresh 

24 

1.8 

1558 

SE  =  3 

interferometric 

0 

5 

16 

doq 

fresh 

26 

4 

&  12 

1559 

2 

interferometric 

0 

2 

23 

rabbit 

5 

1566    1560  -  1571 

2 

24 

Norma 

Spleen 

pig 

fresh 

24 

1.8 

1578 

SE  =  3 

interferometric 

0 

5 

16 

pig 

fresh 

c .  b 

1515 

SD  =  15.2 

4 

SI ng-around 

u 

c 
D 

09 

LL 

cow 

fresh 

24 

1.8 

1577 

SE  =  2 

interferometric 

0 

5 

16 

cow 

fresh 

24 

1.8 

1578 

SE  =  3 

— 

interferometric 

0 

5 

16 

horse 

fresh 

24 

1.8 

1595 

SE  =  3 

interferometric 

0 

5 

16 

dog 

fresh 

26 

4 

&  12 

1570 

2 

interferometric 

0 

2 

23 

Normal  Lung 

dog 

fresh 

35.0  ± 

.5 

0.98 

650 

0 

2 

ref lecti  on 
coefficient^ 

25 

dog 

fresh 

35.0  ± 

.5 

1 
5 

658 
812 
1180 

— 

reflection 
coef f i  ci  ent^ 

less  than 
10  ms'i 

52 

doa 

0.  39 
1 

300 
580 

pulse 

55 

Abnormal  Lung 

dog,  pneu- 
monitic 

fresh 

35 

0.98 

340 

0 

1 

reflection 
coefficient^ 

25 

Normal  Connective  Tissue 

man,  breast 

fresh/dry 

25 

2.0 

1545 
Norma' 

Muscle 

pul  se 

0. 

2 

20 

Skeletal  muscle:  parallel 

to  muscle 

f i  bres 

dog 

fresh 

26 

4 

&  12 

1592 

SD  =  5.0 

7 

interferometric 

0 

2 

23 

rabbit 

fresh 

26 

4 

&  12 

1603 

SD  ^  7.9 

13 

interferometric 

0 

2 

23 

345 


Species        Condition      Ternp.        f  c      Variation        N  Technique        Precision  Source 

(°C)       (MHz)      (m/s)       (m/s)  (%) 


Skeletal  muscle:    perpendicular  to  muscle  fibres 


man,  external 

fresh 

20 

6 

14 

1612 

SD 

12.5 

83 

pul  se 

1 

5 

10 

eye 

37 

1631 

SD 

_ 

15.3 

13 

man,  breast 

fresh/dry 

25 

2 

0 

1545 

pul  se 

0 

2 

20 

dog 

fresh 

26 

4 

& 

12 

1576 

SD 

- 

9.6 

8 

i  nterferonietri  c 

0 

2 

23 

rabbi  t 

fresh 

26 

4 

& 

12 

1587 

SD 

5.5 

18 

i  nter f erometri  c 

0 

2 

23 

Skeletal  muscle 

:  direction 

unspecified 

man,  pectoral 

ref ri  gerated 

24 

1 

8 

1568 

SE 

= 

5.5 

-- 

interferometric 

0 

5 

16 

man 

-- 

24 

1 

8 

1585 

SE 

= 

6 

interferometric 

0 

5 

16 

pig 

fresh 

24 

1 

8 

1580 

SE 

3 

interferometric 

0 

5 

16 

cow 

fresh 

C  H 

1 

8 

1581 

SE 

4 

i  nterf erometri  c 

0 

5 

16 

cow 

fresh 

9/1 

1 
1 

8 

1580 

SE 

4 



interferometric 

0 

5 

16 

cow 

refrigerated 

25 

2 

5 

1580 

SD 

= 

16 

12 

sing-around 

0 

5 

22 

horse 

fresh 

24 

1. 

8 

1598 

SE 

4 

_  _ 

interferometric 

0 

5 

16 

Cardiac  muscle 

pig 

fresh 

24 

1 

8 

1587 

SE 

3 

i  nterf erometri  c 

0 

5 

16 

cow 

f  re  s  h 

24 

1 

8 

1584 

SE 

3 

"i  rThprfpyrirnptri  c 

0 

5 

16 

cow 

fresh 

24 

1 

8 

1570 

SE 

2 

interferometric 

n 
u 

c 

J 

1  F. 
10 

horse 

fresh 

24 

1 

8 

1584 

SE 

3 

*  ~ 

interferometric 

0 

5 

16 

dog 

fresh 

26 

4 

& 

12 

1572 

3 

interferometric 

0 

2 

23 

Normal  Nervous  Tissues 

man,  brain 

fresh 

24 

5 

1524 

pul  se 

0 

1 

26" 

(term  foetus) 

24 

1521 

il 

1540 

man,  brain 



1525 

27 

(foetus ) 

man,  brain 

24 

1 

8 

1564 

SE 

4 

— 

interferometric 

0 

5 

16 

pig,  brain 

fresh 

24 

1 

8 

1565 

SE 

2 

i  nterferometric 

0 

5 

16 

pig,  brain 

fresh 

25 

2 

5 

1506 

SD 

15 

7 

interferometric 

0 

5 

22 

cow,  brain 

fresh 

24 

1 

8 

1560 

SE 

2 

i  nterf erometri  c 

0 

5 

16 

cow,  brain 

fresh 

9/1 

1 

8 

1569 

SE 

3 

inter  icruiMcLi  \l 

n 

J 

1  u 

horse,  brain 

fresh 

24 

1 

8 

1560 

SE 

2 

IMLtri  IcrUllltrLi  IL 

I  u 

dog,  brain 

fresh 

25 

2 

5 

1515 

SD 

15 

5 

si  ng-around 

U 

r 

b 

99 
LC 

cat,  brain 

1 i  vi  ng/f resh 

24 

4 

2 

1557 

pul  se 

- " 

11 

37 

1570 

60 

1574 

rabbit,  brain 

-- 

5 

1508 

1380  - 

1570 

3 

24 

man,  optic 

fresh 

20 

6 

->• 

14 

1644 

SD 

25.4 

30 

pulse 

1 

5 

10 

nerve 

37 

1615 

SD 

3.1 

13 

man,  cerebro- 

fresh 

21.8 

2 

1499 

30 

11 

sing-around 

0 

1 

19 

spinal  fluid 

24.4 

1515 

45 

9 

25.0 

1509 

5 

7 

.5 

11 

Abnormal  Nervous 

Tissues 

man, 

fixed  3  h 

19.0 

2 

1524 

2 

6 

.1 

20 

sing-around 

0. 

1 

19 

meningioma 

48  h 

19.8 

1524 

5 

7 

.5 

20 

man , 

19.7 

2 

1557 

31 

20 

sing-around 

0. 

1 

19 

meningioma 

1546 

15 

20 

(5  sections 

1569 

31 

21 

of  1  tumor) 

1548 

31 

14 

1569 

39 

14 

man,  glioma 

fresh 

22.2 

2 

1529 

1 

9 

.1 

20 

sing-around 

0. 

1 

19 

man,  glioma 

fixed 

22.3 

2 

1500 

45 

17 

si  ng--around 

0. 

1 

19 

346 


Species  Condition  Temp.  f  c  Variation  N  Technique  Precision  Source 
 (°C)       (MHz)      (m/s)       (m/s)  (%) 


man , 

fresh 

27.5 

2 

1545 

4  6.2 

41 

sing-around 

0.1 

19 

astrocytoma 

19 

man , 

T 1  xeci 

2 

1517 

121 

27 

<^  1  n  n  -  ri  mi  in  H 

0.1 

astrocytoma 

man , 

f  i  xed 

20.0 

2 

1501 

45 

18 

sing-around 

0.1 

19 

ependymoma 

Normal  Ophthalmic  Tissues 

man,  lens 

in  vivo 

body 

-- 

1585 

±  74b 

53 

— 

— 

28 

man,  lens 

fresh 

34.1 

4 

1641 

SD  =  16^ 

7 

i  nterferometric 

__ 

29 

(mean ) 

man,  lens 

fresh 

37 

4 

i  nterferometri  c 

0.6 

30 

(autopsy) 

1641 

0  SD  =  1.3 

35 

(operative) 

1638 

4  SD  =  3.0 

12 

(autopsy  +  operative) 

1640 

5  SD  =  1.2 

47 

pig,  lens 

fresh 

23 

4 

i  OD  J 

=  fie 

interferometric 

29 

1  n 
JU 

1673 

SD  =  56 

10 

35 

1677 

SD  =  3^ 

10 

cow,  lens 

fresh 

22 

4 

iODU 

1 nterferometri  c 

0.5 

31 

cow,  lens 

refrigerated 

26-31 

5 

ID  iO 

iOUU  iOOU 

i  nterferometri  c 

1 

12 

rabbit  lens 

5 

i  J  H-U 

OU  DO 

R 

J 

24 

man,  vitreous 

in  vivo 

body 

7 

lb44 

±  ii 

c  0 

bi 

28 

man,  vitreous 

fresh 

35.2(mean)  4 

1530 

SD  =  5'= 

10 

interferometric 

-- 

29 

man,  vitreous 

fresh 

3/ 

4 

i  n  ter f erome  tr i  c 

0.6 

30'- 

(autopsy) 

1532 

SD  =  0.6 

35 

(operative) 

1531. 

7  SD  =  0.9 

14 

(autopsy  +  operative) 

T  C  0  0 

A   c  n  —  n  c 
4  oU  -  (J .  b 

4y 

pig,  vitreous 

fresh 

23 

4 

1510 

SD  =  3^ 

10 

interferometric 

-- 

29 

30 

1522 

SD  =  3^ 

9 

35 

1531 

SD  =  3e 

10 

cow,  vitreous 

fresh 

22 

4 

1495 

interferometric 

0.5 

31 

cow,  vitreous 

refrigerated 

26-31 

5 

1516 

1490  -  1544^ 

interferometric 

1 

12 

rabbit. 

5 

1472 

SD  =  16 

6 

— 

24 

vitreous 

vitreous  +  hyal uronidase 

1494 

SD  =  11 

6 

cow,  aqueous 

fresh 

22 

4 

1495 

interferometric 

0.5 

31 

cow,  aqueous 

refrigerated 

26-31 

5 

1497 

1481  -  1525^ 

12 

man,  cornea/ 

in  vivo 

body 

1502 

±  45b 

53 

28 

aqueous 

cow,  cornea 

fresh 

22 

4 

1550 

interferometric 

C.5 

31 

cow,  sclera 

fresh 

22 

4 

1630 

interferometric 

0.5 

31 

Abnormal  Ophthalmic  Tissues 

man,  lens- 

fresh 

35.8 

4 

1643 

1 

interferometric 

— 

29 

cataractous 

1688 

1 

Normal  Bone 

Long  bones:  a 

long  axis 

cow,  phalanx 

fresh 

5 

4030 

SD  =  110 

252 

pulse 

SD  3 

32 

cow,  phalanx 

dried 

5 

4360 

SD  =  170 

120 

pul  se 

SD  3 

32 

cow,  femur 

dried 

5 

4060 

SD  =  40 

144 

pul  se 

SD  -x.  3 

32 

guinea  pig. 

.1 

3158 

2870  -  35419 

40 

pul  se 

— 

33 

femur 

Long  bones:    across  axis 

cow,  phalanx 

fresh 

5 

3160 

SD  =  170 

252 

pulse 

SD  -v-  3 

32 

cow,  phalanx 

dried 

5 

3270 

SD  =  160 

120 

pulse 

SD  3 

32 

cow,  femur 

dried 

5 

3420 

SD  =  340 

144 

pul  se 

SD  -x.  3 

32 

347 


Species        Condition  Temp. 

(°C) 


f  c  Variation 

(MHz)      (m/s)  (m/s) 


Technique 


Precision  Source 

(%) 


Long  bones:    direction  unspecified 

man,  cortical    in  vivo  body 
bone 


5.0  3406 


fresh 
fixed 


dog,  tibia 
horse 

Skull  bone 
man 

man 

outer  layer 
diploe  layer 
inner  layer 


guinea  pig,  fresh 
completely  healed 
broken  femur 
partially  healed 
broken  femur 


refrigerated  22 


body 


&  5    --  3210 
3700 

0.8  3360 
._h 

2920  SD 
3198  SD 
3098  SD 


SD  =  126 

36209 


18 


160 

93 

220 


23 
9 
5 
6 


man,  pre- 
menopause 


in  vTvo 


body 


Abnormal  Bone 
0.1     2968    2788  -  33719  15 

2551    2442  -  27249  I6 


Normal  Breast  Tissue 
2.0     1510    1450  -  1570J    ^  110  pulse 


pulse 

pulse 
pulse 


continuous  wave 
reflection 

pulse 


pulse 


post-menopause 


1468  1430  -  1520J 
Abnormal  Breast  Tissue 


'x-  40 


brain 
17  weeks 
(2  samples) 


28  weeks 
(2  samples) 


40  weeks 
(2  samples) 


24 
24 
37 
37 

24 
24 
37 
37 

24 
24 
37 
37 


1490 
1495 
1520 
1523 

1498 
1502 
1528 
1529 

1521 
1524 
1540 
1540 


6 


0.2 


34 

35 
36 


37 
38  i 


33 


20^ 


man,  carcinoma 

refrigerated 

24 

1 

8 

1573 

SE  =  7 

interferometric 

0 

5 

16 
20^ 

man,  carcinoma 

in  vivo 

body 

2 

0 

1478 

SD  =  28 

17 

pulse 

0 

2 

man,  fibro- 
cystic 

in  vivo 

body 

2 

0 

1531 

SD  =  23 

28 

pul  se 

0 

2 

20'' 

man,  fibro- 
adenosis 

in  vivo 

body 

2 

0 

1529 

SD  =  30 

6 

pul  se 

0 

2 

20^ 

man,  fibro 
adenoma 

in  vivo 

body 

2 

0 

1529 

SD  =  21 

6 

pul  se 

0 

2 

20^ 

Obstetric  Tissues 

man,  cervix- 
pregnant 
nonpregnant 

in  vivo 

body 

5 

1525 
1633 

SD  =  1.63 
SD  =  2.86 

108 
29 

pulse 

006 

39 

man,  amniotic 
fluid 

25 

5 

1510 

40 

man,  milk 

30 

2 

1540 

pul  se 

0 

2 

20 

man,  foetal 

fresh 

5 

pul  se 

0 

1 

26^ 

348 


Species 


Attenuation  of  ultrasound  in  various  tissues  under  various  conditions. 
Condition       Temp.         f  a       Variation        N  Technique       Precision  Source 


Temp. 

(°C) 


f  a.  Variation 

(MHz)      (m-i)  (m'M 


Norma 

D  1  ooci 

58 

1.09 

-- 

-- 

-- 

— 

i! 

0 

1.73 

1. 

8 

3.22 

3. 

0 

6. 33 

4'. 

8 

12!l 

DO 

1.15 

1. 

0 

1.78 

1. 

8 

3.45 

3. 

0 

6.91 

4. 

8 

12.7 

Normal  Fat 

5. 

gHl 

230 

SD  =  22 

9  -  13 

broadband 

— 

6. 

2 

220 

SD  =  15 

7. 

5 

330 

SD  =  24 

7. 

7 

250 

SD  =  18 

8. 

4 

320 

SD  =  26 

8. 

8 

360 

SD  =  15 

9. 

8 

430 

SD  =  20 

9. 

8 

450 

SD  =  24 

10 

46 

500 

SD  =  30 

13. 

9 

D/U 

c  n    -  OQ 

0 

180P 

-- 

bjJcL-Lr  Ulll    ailu  lj'->  '  J 

4. 

4- 

Ul     a    ur  UClUUaMU 

1  7 

360 

n  1 1 1  c 
|J  U  1  b  tr 

1 

10 

typi  cally 

at  least 

spectrum 

5m"  1 

2 

22 

SD  =  3 

6  per 

a  1 1  a  1  y  J  1  J   u  1 

3 

35 

point: 

a  broadband 

4 

47 

pul  se'^ 

5 

63 

1  point 

6 

84 

per  100 

7 

115 

kHz 

8 

5 



i 

6.9 

SD  =  2.3 

narrow  band 

.  25 

3 

18 

SD  =  2.3 

5 

26 

SD  =  8.1 

1 

r 
D 

6.9 

-- 

2 

8 

21 

4 

40 

6 

56 

7 

75 

87 

4.6 

10% 

1 

7 

8.7 

3 

4 

16 

Norma' 

Liver 

1 
1 

0 1 

20.3 

900 

broadband 

1 

13.8 

typically 

at  least  spectrum  analysis 

5m"  1 

2 

27.6 

SD  =  6 

6  per 

of  a  broadband 

3 

43 

point: 

pulse" 

A 
't 

56 

5 

69 

1  point 

6 

81 

per  100 

7 

93 

kHz 

1 

5 

16 

2 

4 

18 

4 

5 

40 

1 

5 

15 

2 

4 

18 

4 

5 

38 

dog  fresh 
(citrated) 


30 


40 


man,  orbit 


fresh 


37 


man,  orbit 


man 


in  vivo 


f  i  xed 


18  ±  2 


man 

man/cow 


pig 


fresh 


fresh 


cow 


man 
man 


(melted) 


in  VIVO 


f  i  xed 


37 


20  -  35 


body 
18  ±  2 


cow 


cow 


fresh 


fresh 


20  -  35 


41 


10 


61 


42 
4 

21 


43 


44 
7 


45 


46 


349 


Species 


Condi  tion 


Temp. 

(°C) 


f 

(MHz) 


(m-i) 


Variation 
(m-i) 


Technique       Precision  Source 


cow 


fresh 


25 


7.7 


+  10%  17 


cow 


cow 


ground  (±  ^  0.5) 
refrigerated 


fixed 


20  -  35 


.35 
.575 
.6 
.7 

.3 
.87 


13 
20 
29 
54 
71 
104 
4 
6 

10 
9 

11 
9. 


10% 


47 


43 


1  7 

1 4 

97 

cow 

2 

12 

— 

-.r 

54 

12 

narrow  band^ 

Normal 

Kidney 

man/mouse 

fresh/frozen/  22  -  31 

96 

1000 

SD  =  86 

7 

-- 

+  25% 

48 

fixed 

222 

5000 

SD  =  760 

4 

cow 

fresh             20  -  35 

1.5 

19 

— 

— 

— 

45 

2.4 

26 

4.5 

51 

cow 

fresh 

1  t; 

i  .  D 

1  Q 

46 

9  A 

CO 

4.5 

53 

cow 

fixed 

.25 

5 

— 

— 

47 

.3 

4 

.  35 

5 

'a 

6 

.5 

7 

.6 

5 

.7 

9 

.8 

7 

pig 



2 

18 

__r 

— 

54 

16 

narrow  band^ 

Norma' 

Spleen 

man 

fixed  18+2 

1 

3.5 

typically 

at 

least 

spectrum 

5  m- 1 

7 

2 

13 

SD  =  4.6 

6  per 

analysis 

3 

26 

point: 

of  a  broadband 

4 

40 

1  point 

pul se" 

5 

56 

per 

100 

6 

74 

kHz 

7 

96 

Normal  Lung 

man 

fresh 

1 

350 

narrow  band 

0.25  dB 

4 

dog 

fresh           35.0  ±  .5 

.98 

470.0 

10.0 

2 

25 

dog 

fresh           35.0  ±  .5 

1 

430 

410  -  470^' 

52 

2.25 

590 

560  -  640J 

5 

1160 

L040  -  1210J 

dog 

fresh              27  ±  2 

2.4 

440 

SE  =  160 

broadband 

0.1  dB 

56t 

5.0 

900 

SE  =  140 

7.4 

1000 

SE  =  130 

dog,  pneu- 
moni  tic 


fresh 


35.0  ±  .5 


Abnormal  Lung 
350 


25 


350 


Species 


Condition 


Temp . 


f 

(MHz) 


(m-i) 


Variation 
(m-i) 


Technique       Precision  Source 


Normal  Connective  Tissues 


man,  skin  fresh 

— 

1 

40 

SD  = 

14 

3 

85 

SD  = 

14 

5 

106 

SD  = 

25 

man/cow, tendon  fresh 

1 

54 

SD  = 

12 

across  grain 

3 

125 

SD  = 

23 

5 

195 

SD  = 

28 

along  grain 

1 

58 

SD  = 

21 

man/cow  fresh 

— 

1 

38 

SD  = 

10 

articular 

3 

81 

SD  = 

9 

capsule 

5 

130 

SD  = 

62 

ma  n  /  rnui                    "fyf^  Q  h 
■Mail/  uuw                   1  1  Coll 

I 

58 

SD  = 

Ij 

cartilage 

3 

144 

SD  = 

23 

5 

220 

SD  = 

62 

cow,  elastic  fresh 



1 

73 

SD  = 

2 

tendon,  across 

3 

188 

SD  = 

2 

grain 

5 

286 

SD  = 

53 

along  grain 

1 

41 

SD  = 

8 

3 

137 

SD  = 

3 

5 

235 

SD  = 

66 

cow,  rectum  fresh 

1 

6.9 

SD  = 

1.7 

wall 

3 

18 

SD  = 

2 

5 

28 

SD  = 

5 

Norma' 

Muscle 

^l/ala^al    miicr'To*      naval  lol 
■jKc  1  c  Lo  1   IIIUbv.,lc>  parallel 

to  muscle  fibres 

man  fresh 

37 

5. 

gm 

180 

SD  = 

18 

6. 

2 

210 

SD  = 

18 

7. 

5 

0/1  n 

SD  = 

15 

7. 

7 

220 

SD  = 

16 

8. 

4 

310 

SD  = 

24 

8. 

8 

320 

SD  = 

25 

9. 

8 

410 

SD  = 

22 

9. 

8 

400 

SD  = 

24 

10. 

46 

440 

SD  = 

20 

13 

9 

590 

SD  = 

17 

man/cow  fresh 

1 

16 

SD  = 

3 

3 

48 

SD  = 

9 

5 

71 

SD  = 

17 

cow 

20  - 

35 

3 

9.0 

87 

18.0 

1. 

7 

25.4 

3 

4 

62.1 

Skeletal  muscle:    perpendicular  to  muscle  fibres 

man  fresh 

37 

5 

gm 

120 

SD  = 

16 

6 

2 

170 

SD  = 

21 

7 

5 

200 

SD  = 

22 

7 

7 

130 

SD  = 

18 

8 

4 

220 

SD  = 

33 

8 

8 

280 

SD  = 

35 

9 

8 

290 

SD  = 

20 

9 

8 

300 

SD  = 

23 

10 

46 

370 

SD  = 

24 

13 

9 

480 

SD  = 

18 

man/cow  fresh 

1 

8 

SD  = 

1 

3 

30 

SD  = 

6 

5 

40 

SD  = 

1 

cow 

20  - 

35 

.3 

7.5 

.87 

5.5 

3 

.4 

26.5 

narrow  band 


0.25  dB  4 


9-13  broadband 


10 


narrow  band 


0.25  dB  4 


10%  43 


9-13  broadband 


10 


narrow  band 


0.25  dB  4 


10%  43 


351 


Species        Condition       Temp.         f  a       Variation        N  Technique       Precision  Source 
  (°C)  (MHz)      (m-i  (m-i) 


Skeletal 

muscle:    direction  unspecified 

man 

__ 

8 

9.5 

-- 

-- 

42 

cow 



2 

-- 

21  - 

1159 

 r 

54 

— 

21  - 

249 

narrow  band^ 

Cardiac 

muscle 

cow 

fresh             20  -  35 

1 

5 

30 

- 

— 

45 

2 

4 

45 

4 

5 

80 

cow 

T  rs  s  n  — 

1 
i 

5 

30 

_. 



46 

0 

L. 

4 

45 

A 

5 

80 

cow 

fixed 

O 

C 
D 

35 

5 

4 

8 

575 

10 

6 

10 

7 

14 

8 

14 

dog 

fresh 

? 
c 

20 

SD  = 

10 

12  per    spectrum  analysis  -- 

62 

4 

32 

SD  = 

14 

point:    of  a  broadband 

6 

50 

SD  = 

14 

1  point  pulsei^ 

8 

66 

SD  = 

14 

per  500 

10 

92 

SD  = 

20 

kHz 

Tongue: 

parallel  to  fibres 

cow 

iresn             cu  -  jd 

1 
1 

5 

23 

45 

2 

4 

32 

4 

5 

65 

Tongue: 

perpendicular  to  fibres 

cow 

fresh             20  -  35 

1 

5 

45 

45 

2 

4 

65 

Tongue:    direction  unspecified 

cow  fresh  1.5  20 

2.4  32 

4.5  62 


dog,  infarcted  fresh  --  2 

cardiac  4 
muscle  6 

8 
10 


Abnormal  Muscle 


0 

SE  = 

10 

7  per 

spectrum  analysis 

20 

SE  = 

10 

point: 

of  a  broadband 

60 

SE  = 

15 

1  point 

pulse" 

115 

SE  = 

15 

per  500 

190 

SE  = 

10 

kHz 

man,  brain 


man,  brain 

pig,  (whole 
brain) 

(whi  te 
matter) 

pig,  brain 


cow,  brain 


fixed 


fresh 


fresh 


fixed 


fresh 


20  -  35 


20  -  35 


.3 

.87 
1.7 
3.4 

1 

2.5 

2.5 

.35 

.6 

.8 

.87 
1.7 
3.4 


Nerve  Tissues 

8.5 
14.0 
18.0 
36.5 


17 
40 

18 

3 
4 
5 

8.5 
14.0 
33.5 


10% 


36  -  51 


16  -  21 


12 
6 


10% 


43 


49 

50 


47 


43 


352 


Species        Condition       Temp.         f  a       Variation        N  Technique       Precision  Source 

(°C)        (MHz)      (m-i)  (m-i) 


dog,  brain        in  vivo 

body 

97 

5.4 

-- 

1% 

41 

LaLi    Urdin  vivu 

37 

2 

28 

50 

—        hrriri  Hhfln  d  — 

U  1            <J      U  1  1  <-! 

11 

rat,  spinal       in  vivo 

30-31.5 

98 

10 

8 

12 

7       narrow  band 

5 

cord 

mouse,  spinal    in  vivo 

2  ±  .1 

1 

1.7 

— 

53 

cord 

10  ±  .  1 

5.0 

28  ±  .  1 

9.5 

Parallel  to  nerve  fibres 

man,  optic  fresh 

37 

5. 

gm 

290 

SO 

14 

9  -  13    broadband             ■  — 

10 

nerve 

6. 

2 

280 

SO 

16 

7. 

5 

430 

SD 

22 

7 

7 

400 

SD 

47 

8. 

4 

600 

SD 

= 

38 

8. 

8 

500 

SD 

38 

g 

8 

620 

SD 

38 

9 

8 

670 

SD 

= 

36 

10 

46 

600 

SD 

= 

21 

13 

9 

740 

SD 

28 

man,  medulla 

20-35 

1 

7 

14 

10% 

43 

oblongata 

3 

4 

34 

cow,  sciatic 

20-35 

3 

4 

40 

nerve 

Perpendicular  to  nerve  fibres 

man,  optic  fresh 

37 

5 

gm 

210 

SD 

= 

20 

9-13  broadband 

10 

nerve 

6 

2 

240 

SD 

= 

20 

7 

5 

310 

SD 

= 

28 

7 

7 

230 

SD 

= 

22 

8 

4 

370 

SD 

= 

43 

8 

8 

350 

SD 

= 

29 

y 

Q 
.  O 

Jdu 

c  n 
oU 

9 

8 

410 

SD 

33 

10 

46 

500 

SD 

= 

29 

13 

.9 

630 

SD 

33 

man,  medulla 

20-35 

1 

7 

21 

-       -  10% 

43 

obi ongata 

3 

4 

46 

cow,  sciatic 

20-35 

3 

4 

55 

10% 

43 

nerve 


Normal  Ophthalmic  Tissues 


man , 

lens 

10° 

92P 

spectrum  analysis  — 

13 

i 

+ 

of  a  broadband 

17 

156 

pul  se'^ 

cow , 

lens 

fresh 

28 

3.25 

64 

6 

4 

12 

cow, 

lens 

22 

10 

230 

58 

13.5 

310 

cow. 

vitreous 

22 

6 

6 

58 

10 

12 

18 

20 

30 

29 

cow. 

aqueous/ 

fresh 

25-28 

30 

33 

9 

large  broadband  15% 

12 

vitreous 

Normal 

Bone 

man , 

skull 

fresh 

.3 

23 

7 

broadband 

51 

.6 

52 

16 

.8 

92 

28 

1.2 

170 

51 

1.6 

320 

96 

1.8 

430 

130 

353 


Species        Condition      Temp.        f  a      Variation        N  Technique        Precision  Source 

(°C)       (MHz)      (m-i)  (m-i) 


2 
3 

25 
5 

530 
780 

160 
230 

man,  skull 

fresh/fixed 

8 

150 

_  _ 

__ 

37 

man,  skull 
outer 
diploe 
inner 

fixed 

_. 

.h 

1919 
1616 
2460 

SD  =  474 
SD  =  293 
SD  =  1250 

24 
9 
5 
6 



25% 

38^ 

man,  skull 

1 

150 

49 

man/cow 

fresh 

1 

144 

narrow  band 

0.25 

dB 

4 

dog,  tibia 

refrigerated  22 

3 

c 
0 

150 
220 

— 

— 

— 

35 

horse 

 q 

1 

2 
4 

43 
86 
5 

250 
920 

460  -  580 

36 

Obstetric  Tissues 

cow,  uterus 

fresh 

1 

19.6 

SD  =  2 

narrow  band 

0.25 

dB 

4 

The  velocity  was  calculated  from  measurements  of 
the  reflection  coefficient  made  with  a  thermo- 
couple probe  in  a  standing  wave  field. 

'^The  authors  do  not  indicate  what  this  figure 
represents . 

"Measurements  were  made  on  samples  from  eyes  with 
a  variety  of  tumors  although  no  pathological 
condition  was  seen  in  the  components  of  the  eye. 

"^Reference  30  contains  information  on  the  variation 
of  velocity  with  time  after  death. 

'Data  for  the  calculation  of  the  standard  deviation 
were  taken  from  the  graphs  in  the  reference. 

''calculation  of  the  standard  deviation  was  not 
possible  since  each  figure  quoted  by  the  author 
is  a  mean  of  16  to  25  measurements. 

""The  authors  do  not  give  individual  measurements 
but  only  quote  this  range. 

The  authors  do  not  specify  the  frequency  used, 
except  that  the  pulse  used  "contained  substantial 
components  in  the  range  3  to  4  MHz." 

'The  "average  value"  for  bone  quoted  by  these 
authors  must  be  disregarded  because  of  the  manner 
in  which  it  was  obtained. 

^Only  the  range  can  be  obtained  from  the  data  which 
are  represented  graphically. 

^This  includes  data  from  U.S.A.  and  Australia,  and 
a  study  comparing  measurements  with  radiographic 
findings. 


Data  used  to  calculate  the  mean  and  standard 
deviation  were  those  marked  "a"  in  tables  II  to  V 
of  reference  20  (i .e. those  breasts  diagnosed  by 
pathology  studies  of  excised  tissue). 

'"These  figures  refer  to  the  dominant  frequency  in 
the  acoustic  pulse  from  a  reflector  in  water,  but 
the  authors  comment  that  this  may  change  by  as 
much  as  12  percent. 

"This  method  gives  the  attenuation  as  a  continuous 
function  of  frequency,  but  the  authors  did  not 
include  diffraction  corrections  necessary  for 
accurate  results. 

°Frequency  resolution  300  kHz. 

^The  attenuation  coefficient  was  assumed  to  be 
linear  with  frequency. 

"^This  author  also  gives  some  indication  of  the  tem- 
perature variation  of  the  attenuation  coefficient. 

r 

Pulse  technique  using  piezoelectric  receiver. 

^Continuous  wave  technique  using  radiation  balance 
receiver. 

^These  authors  also  give  information  on  the  vari- 
ation of  the  attenuation  coefficient  with  the 
degree  of  inflation  of  the  lung. 

'^Reference  26  contains  further  information  on  the 
variation  of  velocity  in  foetal  brain  with  the 
gestational  age  of  the  foetus. 


354 


oSpleen      oCardiac  muscle 
o  Fat         Oliver        "Kidney      •Skeletal  muscle 
1700p     I  1    r  1   I  1    (  1  I  


I  . 

8 


1650 


1600 


1550 


1500 


1450- 


1400L     I  I  I   I       I       I   I  1       I   I       I       I   I  L 


o 

•  o 


■  Sclera 

o  Brain     o  Vitreous   o  Cornea 
•  CSF        •Lens  •Aqueous 
 1  I  


n  r 


J  1  I  L 


_    I   I  I  I 

1231231       231231       2345634534  5 

Frequency  (MHz) 


Fig.  1.    Frequency  dependence  of  the  velocity  of  ultrasound  in  various  normal  tissues  at  23  ±  3  °C. 


Spleen          Cardiac  muscle 
Fat             Liver         Kidney          Skeletal  muscle 
1700p     I  1    I  1    I  1    I  1  I  


1650 


1600 


^  1550 


1500 


1450 


1400 


Sclera 

o  Brain           oVitreous  Cornea 
CSF              •Lens  Aqueous 
 1    I  I   I  1 


I  L 


I  I  I  I  \  I   I  L 


I      I       I  I 


J   I  L 


(  L 


1       231      2      31       231       231      2       3      4       5634      534  5 

Frequency  (MHz) 


Fig.  2.    Frequency  dependence  of  the  velocity  of  ultrasound  in  various  normal  tissues  at  36  ±  1  °C. 


355 


100 


■  Buschmann  et  al .  [10] 
— Chivers  and  Hill  [7] 

0  Pohlman  [42] 

1  Dussik  et  al .  [4] 
a  Schwan  et  al .  [21  ] 

•  Colombati  and  Petralia  [43] 
Q Goldman  and  Hueter  [1] 
(pooled  data) 


■  / 


Fig.  3.    Attenuation  of  ultrasound  in  normal  fat. 


0.1 


Frequency  (MHz) 


100 


100 


♦  Mountford  and  Wells  [44] 
—  Chivers  and  Hill  [7] 

V  Hueter  [45] 
T  Hueter  and  Pohlman  [46] 
o  Pauly  and  Schwan  [17] 
iEsche  [47] 

•  Colombati  and  Petralia  [43] 
C3 Goldman  and  Hueter  [1] 

(pooled  data) 


Fig.  4.    Attenuation  of  ultrasound  in  normal  liver. 


1|_ 

0.1 


lOOr 


100 


Frequency  (MHz) 


Kidney:   i.  Kessler  [48] 

7  Hueter  [45] 

T  Hueter  and  Pohlman  [46] 

i  Esche  [47] 
Spleen:  —Chivers  and  Hill  [7] 
Uterus:  •  Dussik  et  al .  [4] 


Fig. 


5.    Attenuation  of  ultrasound  in  normal  abdominal 
organ  tissue  (other  than  liver). 


0.1 


Frequency  (MHz) 


100 


356 


100 


Fig.  6.    Attenuation  of  ultrasound  in  normal  muscle. 


O  tfn 


Skeletal:    along  •  Buschmann  et  al .  [10] 
across  a 

along  ♦Dussik  et  al .  [4] 
across  o 

along  •Colombati  and  Petralia  [43] 
across  o 

oPohlman  [42] 
Cardiac:  Allueter  [45] 

AEsche  [47] 
Tongue:       along   THueter  [45] 
across  v 

CjGoldman  and  Hueter  [1] 
(pooled  data) 


Frequency  (MHz) 


100 


(Man) 
(Cow) 


oColombati  and  Petralia  [43] 


(White  matter)  v 


iBallantine  et  al .  [49] 
'Hueter  and  Bolt  [50] 


□  Vosioka  et  al .  [41] 
^Esche  [47] 

I  Robinson  and  Lei e  [11] 
QGoldman  and  Hueter  [1] 
(pooled  data) 


Fig,  7.    Attenuation  of  ultrasound  in  normal  brain. 


iL_ 

0.1 


lOOr 


Frequency  (MHz) 


□  0 

o  a 


100 


Medulla  oblongata: 
Spinal  cord: 


along  DBuschmann  et  al.  [10] 
across  ■ 

along  oColombati  and  Petralia  [43] 
across  ♦ 

along  OColombati  and  Petralia  [43] 
across  • 

vFry  and  Fry  [5] 


Fig.  8.    Attenuation  of  ultrasound  in  normal  nervous 
tissues  (other  than  brain). 


0.1 


100 


Frequency  (MHz) 


357 


1000 


vDussik  et  al.  [4] 
oAdler  and  Cook  [35] 
□  Kishimoto  [36] 

Skull  bone 

•  Hueter  [51] 

■  Theismann  and  Plander  [37] 
iBallantine  et  al .  [49] 


100 


Fig.  9.    Attenuation  of  ultrasound  in  normal  bone. 


10  100 

Frequency  (MHz) 


References 

[1]    Goldman,  D.  E.  and  Hueter,  T.  F.,  Tabular 
data  of  the  velocity  and  absorption  of  high- 
frequency  sound  in  mammalian  tissues,  J. 
Acoust.  Soc.  Am.  28,  35-37  (1956). 
Errata:    J.  Acoust.  Soc.  Am.  29,  655  (1957). 

[2]  Wells,  P.  N.  T.,  Absorption  and  dispersion 
of  ultrasound  in  biological  tissue.  Ultra- 
sound in  Med.  Biol.  1,  369-376  (1975y^ 

[3]    Hussey,  M. ,  Diagnostic  ultrasound  (Blackie, 
Glasgow,  1975). 

[4]    Dussik,  K.  T.,  Fritch,  D.  J.,  Kyriazidou,  M. , 
and  Sears,  R.  S.,  Measurements  of  articular 
tissues  with  ultrasound.  Am.  J.  Phys.  Med. 
3^,  160-165  (1958). 

[5]    Fry,  W.  J.  and  Fry,  R.  B.,  Temperature 

changes  produced  in  tissue  during  ultrasonic 
irradiation,  J.  Acoust.  Soc.  Am.  25 ,  6-11 
(1953). 

[6]    Zieniuk,  J.  and  Chivers,  R.  C,  Measurement 
of  ultrasonic  exposure  with  radiation  force 
and  thermal  methods,  Ultrasonics  14,  161- 
172  (1976). 

[7]  Chivers,  R.  C.  and  Hill,  C.  R.,  Ultrasonic 
attenuation  in  human  tissue,  Ultrasound  in 
Med.  Biol.  2,  25-29  (1975). 

[8]    McSkimin,  H.  J.,  Ultrasonic  Methods  for 
Measuring  the  Mechanical  Properties  of 
Liquids  and  Solids  in  Physical  Acoustics, 
W.  P.  Mason,  ed. ,  Vol.  1,  Part  A,  pp.  271- 
334  (Academic  Press,  New  York,  1964). 

[9]    Wehr,  J.  S.,  Classification  and  discussion 
of  interference  methods  for  measuring  the 
propagation  velocity  of  ultrasonic  waves,  in 
Proceedings  of  the  Fourth  International 
Conference  on  Acoustics,  Copenhagen,  Paper 
K15  (1962). 


[10]    Buschmann,  W.,  Voss,  M.  ,  and  Kemmerling,  S., 
Acoustic  properties  of  normal  human  orbit 
tissues,  Ophthal.  Res.  1,  354-364  (1970). 

[11]    Robinson,  T.  C.  and  Lele,  P.  P.,  An  analysis 
of  lesion  development  in  the  brain  and  in 
plastics  by  high-intensity  focussed  ultra- 
sound at  low-megahertz  frequencies,  J. 
Acoust.  Soc.  Am.  5^,  1333-1351  (1972T. 

[12]    Begui,  Z.  E. ,  Acoustic  properties  of  the 

refractive  media  of  the  eye,  J.  Acoust.  Soc. 
Am.  2i,  365-368  (1954). 

[13]    Lizzi,  F.,  Katz,  L. ,  St.  Louis,  L. ,  and  Cole- 
man, D.  J.,  Applications  of  spectral  analysis 
in  medical  ultrasonography.  Ultrasonics  14, 
77-80  (1976). 

[14]    Wehr,  J.  S.,  Error  analysis  as  a  factor  in 
the  choice  of  the  proper  method  of  measur- 
ing the  attenuation  and  the  reflection  coef- 
ficient of  ultrasonic  waves  in  solids,  in 
Proceedings  of  the  Conference  on  Acoustics 
of  Solid  Media  (PWN  Warsaw,  1964),  pp.  239- 
245. 

[15]    Frucht,  A.  H. ,  Die  Geschwindigkeit  des  Ultra- 
schalls  in  menschlichen  und  tierischen  Ge- 
weben,  Naturwi ssenschaf ten  39,  491-492 
(1952). 

[16]    Frucht,  A.  H.,  Die  Schallgeschwindigkeit  in 
menschlichen  und  tierischen  Geweben,  Ze i t . 
Ges.  Exptl.  Med.  120,  526-557  (1953). 

[17]  Pauly,  H.  and  Schwan,  H.  P.,  Mechanism  of 
absorption  of  ultrasound  in  liver  tissue, 
J.  Acoust.  Soc.  Am.  50,  692-699  (1971). 

[18]    Danckwerts,  H.  J.,  Discrete  relaxation 

processes  as  a  model  of  the  absorption  of 
ultrasound  in  liver  homogenate,  J.  Acoust. 
Soc.  Am.  55_,  1098-1099  (1974). 


358 


[19]    Van  Venrooij,  G.  E.  P.  M. ,  Measurement  of 
ultrasound  velocity  in  human  tissue.  Ultra- 
sonics 9.  240-242  (1971). 

[20]    Kossoff,  G.,  Fry,  E.  K. ,  and  Jellins,  J., 
Average  velocity  of  ultrasound  in  the  human 
female  breast,  J.  Acoust.  Soc.  Am.  53, 

1730-1736  (1973y: 

[21]    Schwan,  H.  P.,  Carstensen,  E.  L.  ,  and  Li,  K. , 
Heating  of  fat-muscle  layers  by  electromag- 
netic and  ultrasonic  diathermy.  Trans.  Am. 
Inst.  Elec.  Engrs.  1,  72,  483-488  (1953). 

[22]    Ludwig,  G.  D. ,  The  velocity  of  sound  through 
tissues  and  the  acoustic  impedance  of  tissues, 
J.  Acoust.  Soc.  Am.  22,  862-866  (1950). 

[23]    Goldman,  D.  E.  and  Richards,  J.  R.,  The  mea- 
surement of  high  frequency  sound  velocity  in 
mammalian  soft  tissue,  J.  Acoust.  Soc.  Am. 
26,  981-984  (1954). 

[24]    Nakajima,  A.,  Nishi,  S.,  Amano,  K. ,  Uesugi , 
T. ,  and  Wakai ,  T. ,  Measurement  of  the  axial 
length  of  the  eye  by  ultrasonic  device,  (in 
Japanese),  J.  Clin.  Ophth.  14,  1594-1599 
(1960). 

[25]    Dunn,  F.  and  Fry,  W.  J.,  Ultrasonic  absorp- 
tion and  reflexion  by  lung  tissue,  Phys.  Med. 
Biol.  5,  401-410  (1961). 

[26]    Wladimiroff,  J.  W.,  Craft,  I.  L.,  and  Talbert, 
D.  G. ,  In  vitro  measurements  of  sound  velocity 
in  human  fetal  brain  tissue.  Ultrasound  in 
Med.  Biol.  U  377-382  (1975). 

[27]    Willocks,  J.,  Donald,  I.,  Duggan,  T.  C,  and 
Day,  N. ,  Foetal  cephalometry  by  ultrasound, 
J.  Obstet.  Gynaec.  Brit.  Cwlth.  71.,  11-20 
(1964). 

[28]    Nakajima,  A.,  Kimura,  T.,  and  Yamazaki,  M., 
Applications  of  Ultrasound  in  Biometry  of 
the  Eye  in  Ultrasonics  in  Ophthalmology, 
R.  E.  Goldberg  and  L.  K.  Sarin,  eds.,  pp. 
124-144  (W.  B.  Saunders,  London  and  New  York, 
1967). 

[29]    Jansson,  F.  and  Sundmark,  E.,  Determination 
of  the  velocity  of  ultrasound  in  ocular  tis- 
sues at  different  temperatures.  Acta  Ophthal . 
39,  899-910  (1961). 

[30]    Jansson,  F.  and  Kock,  E.,  Determination  of 
the  velocity  of  ultrasound  in  the  human  lens 
and  vitreous.  Acta  Ophtal .  40,  420-433  (1962). 

[31]    Oksala,  A.  and  Lehtinen,  A.,  Measurement  of 
the  velocity  of  sound  in  some  parts  of  the 
eye.  Acta  Ophtal.  36,  633-639  (1958). 

[32]    Lang,  S.  B.,  Ultrasonic  method  for  measuring 
elastic  coefficients  of  bone  and  results  on 
fresh  and  dried  bovine  bones,  I.E.E.E.  Trans. 
Bio-Med.  Eng.  BME-17,  101-105  (1970). 

[33]    Floriani,  L.  P.,  Debevoise,  N.  T.,  and  Hyatt, 
G.  W.  ,  Mechanical  properties  of  healing  bone 
by  the  use  of  ultrasound,  Surg.  Forum  18,  468- 
470  (1967). 


[34]    Craven,  D.  A.,  Constantini,  M.  A.,  Greenfield, 
M.  A.,  and  Stern,  R.,  Measurement  of  the 
velocity  of  ultrasound  in  human  cortical  bone 
and  its  potential  clinical  importance.  An 
in  vivo  preliminary  study.  Invest.  Radiol.  8, 
72-77  (1973). 

[35]    Adler,  L.  and  Cook,  K.  V.,  Ultrasonic  para- 
meters of  freshly  frozen  dog  tibia,  J.  Acoust. 
Soc.  Am.  58,  1107-1108  (1975). 

[36]    Kishimoto,  T. ,  Ultrasonic  absorption  in  bones, 
Acustica  8,  179-180  (1958). 

[37]  Theismann,  H.  and  Pfander,  F. ,  Uber  die 
DurchlHssigkeit  des  Knochens  fUr  Ultra- 
schall,  Strahlentherapie  80,  607-610  (1949). 

[38]    Martin,  B.  and  McElhaney,  J.  H.,  The 

acoustic  properties  of  human  skull  bone, 
J.  Biomed.  Mater.  Res.  5,  325-333  (1971). 

[39]    Bakke,  T.  and  Gytre,  T. ,  Ultrasonic  measure- 
ment of  the  sound  velocity  in  the  pregnant 
and  the  non-pregnant  cervix  uteri,  Scand. 
J.  Clin.  Lab.  Invest    33,  341-346  (1974). 

[40]    Zana,  R.  and  Lang,  J.,  Interaction  of  ultra- 
sound with  amniotic  liquid,  J.  Acoust.  Soc. 
Am.  55,  391  (1974). 

[41]    Yosioka,  K. ,  Ohmura,  A.,  Hasegawa,  T. ,  and 
Oka,  M.,  Absorption  coefficient  of  ultra- 
sound in  soft  tissues  and  their  biological 
conditions,  in  Proceedings  of  the  6th  Inter- 
national Congress  on  Acoustics,  Tokyo,  1968, 
pp.  M5-M8. 

[42]    Pohlman,  R.,  Uber  die  Absorption  des  Ultra- 
schalls  im  menschlichen  Gewebe  und  ihre  Ab- 
hclngigkeit  von  der  Frequenz,  Physi kal i sche 
Zeit.  40,  159-161  (1939). 

[43]    Colombati,  S.  and  Petralia  S. ,  Assorbimento 
di  ultrasuoni  in  tessuti  animal i.  La  Ricerca 
Scientifica  20,  71-78  (1950). 

[44]    Mountford,  R.  A.  and  Wells,  P.  N.  T. ,  Ultra- 
sonic liver  scanning:    the  quantitative 
analysis  of  the  normal  A-scan,  Phys.  Med. 
Biol .  17_,  14-25  (1972). 

[45]    Hueter,  T.  F.,  Messung  der  Ul traschal 1  absorp- 
tion in  tierischen  Geweben  und  ihre  Abh^ngig- 
keit  von  der  Frequenz,  Naturwi ssenschaften 
35,  285-287  (1948). 

[46]    Hueter,  T.  F.  and  Pohlman,  R. ,  Eine  optische 

Methode  zur  Bestimmung  der  Ultraschal labsorp- 

tion  in  undurchsichtigen ,  schal Iwei chen 

Medien,  Z.  Angew.  Phys.  I,  405-411  (1949). 

[47]    Esche,  R.,  Untersuchungen  zur  Ultraschall- 
absorption  in  tierischen  Geweben  und  Kunst- 
stoffen,  Akust.  Beih.  2,  71-74  (1952). 

[48]    Kessler,  L.  W.,  VHF  ultrasonic  attenuation 
in  mammalian  tissue,  J.  Acoust.  Soc.  Am.  53, 
1759-1760  (1973). 

[49]    Ballantine,  H.  T.,  Hueter,  T.  F.,  and  Bolt, 
R.  H.,  On  the  use  of  ultrasound  for  tumour 


359 


detection,  J.  Acoust.  Soc.  Am.  26,  581  (1954). 

[50]    Hueter,  T.  F.  and  Bolt,  R.  H.,  An  ultrasonic 
method  for  outlining  the  cerebral  ventricles, 
J.  Acoust.  Soc.  Am.  23,  160-167  (1951). 

[51]    Hueter,  T.  F.,  Messung  der  Ul traschal 1  absorp- 
tion im  menschlichen  SchSdel knochen  und  ihre 
AbhSngigkeit  von  der  Frequenz,  Naturwissen- 
schaften  39,  21-22  (1952). 

[52]  I  Dunn,  F.,  Attenuation  and  speed  of  ultrasound 
in  lung,  J.  Acoust.  Soc.  Am.  56,  1638-1639 

(1974)  . 

[53]    Dunn,  F.,  Temperature  and  amplitude  depend- 
ence of  acoustic  absorption  in  tissue,  J^. 
Acoust.  Soc.  Am.  34,  1545-1547  (1962). 

[54]  Marcus,  P.  W.  and  Carstensen,  E.  L.,  Problems 
with  absorption  measurements  of  inhomogeneous 
solids,  J.  Acoust.  Soc.  Am.  58,  1334-1335 

(1975)  . 

[55]    Reid,  J.  M.,  (1965),  reported  in  F.  Dunn, 

P.  D.  Edwards  and  W.  J.  Fry,  Absorption  and 
Dispersion  of  Ultrasound  in  Biological  Media, 
H.  P.  Schwan  ed.,  pp.  205-332  (McGraw-Hill 
Book  Co.,  New  York,  1969). 

[56]    Bauld,  T.  J.  and  Schwan,  H.  P.,  Attenuation 

and  reflection  of  ultrasound  in  canine  lung 

tissue,  J.  Acoust.  Soc.  Am.  56,  1630-1637 
(1974). 


[57]    Carstensen,  E.  L.,  Li,  K. ,  and  Schwan,  H.  P., 
Determination  of  the  acoustic  properties  of 
blood  and  its  components,  J.  Acoust.  Soc.  Am. 
25,  286-289  (1953). 

[58]    Filipczynski ,  L.,  Etienne,  J.,  Lypacewicz,  G., 
and  Salkowski,  J.,  Visualizing  internal  struc- 
tures of  the  eye  by  means  of  ultrasonics,  Proc. 
Vibration  Problems  8,  357-368  (1967). 

[59]    Urick,  R.  J.,  A  sound  velocity  measurement  for 
determining  the  compressibility  of  finely  di- 
vided substances,  J.  Appl.  Phys.  18^,  983-987 
(1947). 

[60]    Heyser,  R.  C.  and  Le  Croissette,  D.  H.,  A  new 
ultrasonic  imaging  technique  using  time  delay 
spectrometry,  Ultrasound  in  Med.  Biol.  U 
119-131  (1975). 

[61]    Lizzi,  F.  L.  and  Laviola,  M.  A.,  Power  spectra 
measurements  of  ultrasonic  backscatter  from 
ocular  tissues,  in  Proceedings  of  1975  I.E.E.E. 
Ultrasonics  Symposium,  pp.  29-32  (1975). 

[62]    Yuhas,  D.  E.,  Mimbs,  J.  W.,  Miller,  J.  G., 
Weiss,  A.  N.,  and  Sobel ,  B.  E.,  Changes  in 
ultrasonic  attenuation  indicative  of  regional 
myocardial  infarction,  in  Proceedings  of 
World  Federation  of  Ultrasound  in  Medicine 
and  Biology,  San  Francisco,  1976,  pp.  1883- 
1894  (1976). 


360 


SUBJECT  INDEX 


Absorption,  19,  29,  43,  153 

Acoustoel ectri c  effect,  63 

Amino  acids,  19 

Anisotropy,  73,  189 

Attenuation,  19,  37,  43,  63,  73,  81, 
85,  93,  101,  121,  125,  153,  157 
197,  203,  209,  247,  255,  281,  287, 

343 

Backscattering,  153,  157,  165,  247, 
275,  327 

Beam  displacement,  203 

Beam  distortion,  203 

Blood,  165,  173 

Bone,  19,  179,  189,  197 

Brain,  81,  203,  209 

Breast,  13-16,  85,  93,  173,  221,  255 

Cell  detachment,  317 

Cepstrum,  287 

Collagen,  19,  73,  179,  189 

Color-coded  B-scan,  255,  261 

Computed  tomography,  227,  235,  247, 
255,  337 

Convolution,  287 

Correlation  analysis,  275 

De-convolution,  287 

Dispersion,  81 ,  189,  197 

Doppler,  173,  227,  235 

Dynamic  autocorrelation  analysis,  275 

Elasticity,  73,  189 

Estimation  theory,  125 

Eye,  19,  111 

Filtering,  299-300 

Hamming  window,  281 

Heart,  43,  73,  153,  267 

Homomorphic  processing,  287 

Hydroxyapatite,  179,  189 

Hyperthermia,  57 

Image  analysis,  255,  303 

Imaging, 
annular  array,  255 
brain  (transkul 1 ) ,  235 
Doppler  (moving  media),  235 
image  reconstruction  (see  computed 

tomography) 
synthetic  focus,  235 

Impedance,  81 

Inhomogenei ty  thermal  losses,  37 
Interferogram,  73 
Kidney,  303 


Lag  windows,  281 
Lipids,  157 

Liver,  125,  153,  157,  287,  303 
Lung,  19,  135,  (see  scattering) 
Microscopy,  myocardium,  73 
Myocardium,  63,  73,  153,  267 
Opto-acoustic  visualization,  255 
Pattern  recognition,  297,  303 
Phantoms,  179,  323,  327,  337 
Phase  compensation,  209,  235 
Polypeptides,  19 

Probability  density  functions,  125 

Prostate,  297 

Proteins,  19,  157 

Pulse  compression,  255 

Relaxation  phenomenon,  29 

Scattering,  37,  111 ,  135,  143,  153, 
157,  281  (see  also  backscattering) 

Sensitivity  enhancement,  255 

Signal  averaging,  255 

Skull ,  197,  203,  209 

Spectral  analysis,  43,  111,  125,  143, 
261  ,  267,  281 ,  287 

Speed,  (see  velocity) 

Temperature  dependence  of  ultrasonic 
properties,  57,  63,  203,  227,  235 

Temporal  changes,  267,  275 

Thermal  "lesions",  203 

Thermal  wave,  37 

Thermodynamics,  189 

Tissue  death,  317 

Tomography,  (see  computed  tomography) 

Transducer  calibration,  255 

Velocity,  19,  43,  53,  57,  63,  73,  81, 
179,  189,  197,  343 

Visco-elasticity,  189 

Viscous  relative  motion,  37 


361 


AUTHOR  INDEX 


Bahn,  R.  C,  227  ,235 
Barger,  J.  E. ,  197 
Barnes,  R.  W. ,  81 
Baxter,  B.  ,  235 
Beaver,  W. ,  267 
Birnholz,  J. ,  287 
Bowen ,  T . ,  57 
Boyle,  D. ,  267 
Boynard,  M. ,  165 
Brady,  J.  K. ,  19 
Busey,  H. ,  323 

Carson,  P.  L. ,  337 
Cartensen ,  E .  L . ,  29 
Christensen,  D. ,  235 
Chivers,  R.  C. ,  343 
dayman,  W.  ,  337 
Connor,  W.  G. ,  57 
Czerwinsk  i ,  M.  J. ,  303 

Dain  P.,  125 
Davidson,  C.  L. ,  179 
Dick,  D.  E. ,  337 
Dietz,  D.  R. ,  255 
Doppman ,  J .  L . ,  255 
Duck,  F.  A. ,  247 
Dunn,  F. ,  19,43 

Edmonds,  P.  D. ,  323 
Eggletoti,  R.  C.  ,  327 
Elbaum,  M.  E. ,  111 

Filly,  R.  A. ,  323 
Finby,  N. ,  125 
Eraser,  J. ,  287 
Freese,  M. ,  157 
Frizzell ,  L.  A. ,  19,43 
Fry,  E.  K. ,  85 
Fry,  F.  J. ,  85,203 

Gaca,  A. ,  297 
Gallagher  H.  S. ,  85 
Gammell ,  P.  M. ,  101 
Glover,  G.  H.,  221 
Gore,  J.  C. ,  275 
Goss,  S.  A. ,  19,43 
Graniiak,  R.  ,  143 
Greenleaf ,  J.  F. ,  227,235 

Halliwell,  M.,  173 
Hanss,  M. ,  165 
Heyser ,  R.  C. ,  101 
Higgins,  F.  P. ,  255 
Hill ,  C.  R. ,  247 
Holasek,  E. ,  261 
Hunter,  L.  P. ,  143 

Jennings,  W.  D. ,  261 
Johnson,  S.  A. ,  227,235 
Johnston,  R.  L.,  19,43 
Joynt,  L. ,  267 


Katz,  J,  L. ,  189 
Kessler,  L.  W. ,  73 
Keuwez,  J. ,  121 
Kino,  G.  S. ,  287 
Kobayashi ,  T. ,  93 
Kremkau,  F.  W. ,  81 
Kuc,  R.,  125 

Leb,  D.  E.,  303 

Le  Croissette,  D.  H. ,  101 

Lee,  P.  P.  K.,  143 

Leeman,  S. ,  275 

Lees,  S.,  179 

Lerner,  R.  M. ,  143 

Levi,  S,,  121 

Linzer ,  M. ,  255 

Lizzi,  F.  L.,  Ill 

Lock,  E.,  297 

Lyons,  E.  A. ,  157 

Maynard,  V. ,  19,43 
McGraw,  C.  P. ,  81 
Metrewel i ,  C. ,  275 
Miller,  J.  G.,  37,63 
Mimbs,  J.  W.,  63 
Mountford,  R.  A. ,  173 

Nasoni ,  R.  L. ,  57 
Nider,  L.,  43 
Norton,  S.  J. ,  255 

O'Brien,  W.  D. ,  Jr. ,  19,43 
O'Donnell,  M.,  37,63 

Parkinson,  D.  B. ,  323 
Parks,  S.  I.,  255 
Parry,  R.  J. .  343 
Phillips,  D.  J.,  209 
Pifer,  A.  E.,  57 
Plessner,  N.  J. ,  275 
Popp,  R. ,  267 
Preston,  K. ,  Jr. ,  303 
Purnell,  E.  W.,  261 

Rakowski,  H.,  267 
Reid,  J.  M.  ,  153 
Reyes,  Z.  ,  323 
Rhyne,  T.  L.,  135 
Robinson,  D.  E. ,  281 
Rajagopalan,  B.,  227,235 
Roseboro,  J.  A. ,  101 


Sanghvi ,  N.  T 
Scheiding,  W. 
Schenk,  E.  A. 
Schwartz,  M. , 
Shabason,  L. , 
Shawker,  T.  H 
Shideler,  R. 
Sholes,  R.  R. 
Shung,  K.  K. , 
Skidmore,  R. , 
Skolnik,  M.  L 
Smith,  S.  W. , 
Sobel,  B.  E., 
Sollish,  B.  D 


.,  85 
,  297 
,  143 
125 
337 

.,  255 
W.,  255 
,  57 
153 
173 

.,  303 

209 
63 

.,  53 


Thomas,  P.  J.,  227 
Thurstone,  F.  L.,  209 

von  Ramm,  0.  T. ,  209 
von  Seelen ,  W. ,  297 

Waag,  R.  C,  143 
Webb,  A.  J.,  173 
Weiss,  L. ,  317 
Wells,  P.  N.  T. ,  173 
Wessels,  G. ,  297 
Whitcomb,  J.  A. ,  327 
Willson,  K. ,  275 
Wilson,  R.  L. ,  101 
Woodcock,  J.  P. ,  173 

Yoon,  H.  S.,  189 
Yuhas,  D.  E. ,  73  ^  • 


362 


NBS-114A  IREV.  0-761 


U.S.  DEPT.  OF  COMM. 

1.  PUBLICATION  OR  REPORT  NO.           l||^t||H|||p|P  tiOj 

BIBLIOGRAPHIC  DATA 
SHEET 

NBS  SP  525  p-..*-^r^-->f^-% 

li      ....       .•  s 

4.  TITLE  AND  SUBTITLE 

Ultrasonic  Tissue  Characterization  II 

5.  Publication  Date 

April  1979 

rmwmi  og  Orgamnioon  Code 

7.  AUTHOR(S) 

Edited  by  Melvin  Linzer 

8.  Performing  Organ,  Report  No. 

9.  PERFORMING  ORGANIZATION  NAME  AND  ADDRESS 

1ft,  Pro|ect/Ta«J(/*<Kt(  Onii  No. 

NATIONAL  BUREAU  OF  STANDARDS 
DEPARTMENT  OF  COMMERCE 
WASHINGTON,  DC  20234 

11.  Contract/Grant  No. 

12.  SPONSORING  ORGANIZATION  NAME  AND  COMPLETE  ADDRESS  fStreef,  city.  State.  ZIP) 

National  Science  Foundation 
Washington,  D.C. 

and  National  Institutes  of  Health 
Bethesda,  Maryland 

13.  Type  of  Report  &  Period  Covered 

14.  Sponsoring  Agency  Code 

b.^.^.....--........:.  :  

15.  SUPrLEMENTARY  NOTES 


Library  of  Congress  Catalog  Card  Number:  79-600026 

I    I  Document  describes  a  computer  program;  SF-185,  FIPS  Software  Summary,  is  attached. 

16.  ABSTRACT  (A  200-word  or  leaa  tactual  aumrnary  of  most  significant  Information.    If  document  includes  a  significant  bibliography  or 
literature  survey,  mention  it  here.) 


The  Second  International  *^vmDosium  on  Ultrasonic  Imaging  and  Tissue 
Characterization  was  held  at  the  National  Bureau  of  Standards  on  June  13-15,  1977. 
The  meeting  was  cosponsored  by  the  National  Bureau  of  Standards,  the  National 
Science  Foundation,  and  the  National  Institutes  of  Health.    This  volume  contains 
extended  and  reviewed  papers  based  on  43  of  the  53  talks  presented  at  the 
Symposium.    Topics  covered  include  techniques  for  measurement  of  ultrasonic 
tissue  parameters,  the  dependence  of  tissue  properties  on  physical  and  biological 
variables  ^e.g.,  ultrasonic  frequency,  temperature),  mechanisms  of  ultrasonic 
tissue  interactions,  propagation  through  bone  and  skull,  tumpr  Doppler  signatures, 
computerized  tomography,  signal  processing  and  pattern  recognition,  and  tissue 
phantoms.    A  survey  of  velocity  and  attenuation  data  in  mammalian  tissue  is  included 
in  an  appendix. 


17.  KEY  WORDS  fs/x  to  twelve  entries;  alphabetical  order;  capitalize  only  the  first  letter  of  the  first  key  word  unless  a  proper  name; 
separated  by  semicolons) 


Absorption;  attenuation;  computerized  tomography;  Doppler;  impedance;  medical 
diagnosis;  microscopy;  pattern  recognition;  scattering;  signal  processing; 
tissue  characterization;  tissue  parameters;  ultrasound;  velocity. 


18.  AVAILABILITY                          |3  Unlimited 

19.  SECURITY  CLASS 
(THIS  REPORT) 

21.  NO.  OF 
PRINTED  PAGES 

1    1  For  Official  Distribution.  Do  Not  Release  to  NTIS 

UNCLASSIFIED 

339 

rrp  Order  From  Sup.  of  Doc,  Government  Printing  Office,  Washington,  DC 
^  20402,  SD  Stocl<  No.  SN003-003-0258-3 

20.  SECURITY  CLASS 
(THIS  PAGE) 

22.  Price 

1    1  Order  From  National  Technical  Information  Service  (NTIS),  Springfield, 
VA.  22161 

UNCLASSIFIED 

$5.50 

ir  U.S.  GOVERNMENT  PRINTING  OFFIfE:  1979    O— 291-026 


I 


i 


I 

I 
i 
i 


I 

1. 

1 
i 


NBS  TECHNICAL  PUBLICATIONS 


PERIODICALS 

JOURNAL  OF  RESEARCH— The  Journal  of  Research 
of  the  National  Bureau  of  Standards  reports  NBS  research 
and  development  in  those  disciplines  of  the  physical  and 
engineering  sciences  in  which  the  Bureau  is  active.  These 
include  physics,  chemistry,  engineering,  mathematics,  and 
computer  sciences.  Papers  cover  a  broad  range  of  subjects, 
with  major  emphasis  on  measurement  methodology,  and 
the  basic  technology  underlying  standardization.  Also  in- 
cluded from  time  to  time  are  survey  articles  on  topics  closely 
related  to  the  Bureau's  technical  and  scientific  programs.  As 
a  special  service  to  subscribers  each  issue  contains  complete 
citations  to  all  recent  NBS  publications  in  NBS  and  non- 
NBS  media.  Issued  six  times  a  year.  Annual  subscription: 
domestic  $17.00;  foreign  $21.25.  Single  copy,  $3.00  domestic; 
$3.75  foreign. 

Note:  The  Journal  was  formerly  published  in  two  sections: 
Section  A  "Physics  and  Chemistry"  and  Section  B  "Mathe- 
matical Sciences." 
I DIMENSIONS/NBS 

\  This  monthly  magazine  is  published  to  inform  scientists, 
I  engineers,  businessmen,  industry,  teachers,  students,  and 
I  consumers  of  the  latest  advances  in  science  and  technology, 

with  primary  emphasis  on  the  work  at  NBS.  The  magazine 
[highlights  and  reviews  such  issues  as  energy  research,  fire 

protection,  building  technology,  metric  conversion,  pollution 
,i  abatement,  health  and  safety,  and  consumer  product  per- 
'{formance.  In  addition,  it  reports  the  results  of  Bureau  pro- 
grams in  measurement  standards  and  techniques,  properties 
jof  matter  and  materials,  engineering  standards  and  services, 
jinstrumentation,  and  automatic  data  processing. 

y   Annual  subscription:  Domestic,  $1 1.00;  Foreign  $13.75 

NONPERIODICALS 

IMonographs — Major  contributions  to  the  technical  liter- 
ature on  various  subjects  related  to  the  Bureau's  scientific 
and  technical  activities. 

•Handbooks — Recommended  codes  of  engineering  and  indus- 
trial practice  (including  safety  codes)  developed  in  coopera- 
tion with  interested  industries,  professional  organizations, 
i  md  regulatory  bodies. 

jipecial  Publications — Include  proceedings  of  conferences 
■ponsored  by  NBS,  NBS  annual  reports,  and  other  special 
publications  appropriate  to  this  grouping  such  as  wall  charts, 
TOcket  cards,  and  bibliographies. 

Applied  Mathematics  Series — Mathematical  tables,  man- 
ijials,  and  studies  of  special  interest  to  physicists,  engineers, 
'  hemists,  biologists,  mathematicians,  computer  programmers, 
nd  others  engaged  in  scientific  and  technical  work, 
''lational  Standard  Reference  Data  Series — Provides  quanti- 
ative  data  on  the  physical  and  chemical  properties  of 
naterials,  compiled  from  the  world's  literature  and  critically 
valuated.  Developed  under  a  world-wide  program  co- 
rdinated  by  NBS.  Program  under  authority  of  National 
tandard  Data  Act  (Public  Law  90-396). 


NOTE:  At  present  the  principal  publication  outlet  for  these 
data  is  the  Journal  of  Physical  and  Chemical  Reference 
Data  (JPCRD)  published  quarterly  for  NBS  by  the  Ameri- 
can Chemical  Society  (ACS)  and  the  American  Institute  of 
Physics  (AIP).  Subscriptions,  reprints,  and  supplements 
available  from  ACS,  1155  Sixteenth  St.  N.W.,  Wash.,  D.C. 
20056. 

Building  Science  Series — Disseminates  technical  information 
developed  at  the  Bureau  on  building  materials,  components, 
systems,  and  whole  structures.  The  series  presents  research 
results,  test  methods,  and  performance  criteria  related  to  the 
structural  and  environmental  functions  and  the  durability 
and  safety  characteristics  of  building  elements  and  systems. 
Technical  Notes — Studies  or  reports  which  are  complete  in 
themselves  but  restrictive  in  their  treatment  of  a  subject. 
Analogous  to  monographs  but  not  so  comprehensive  in 
scope  or  definitive  in  treatment  of  the  subject  area.  Often 
serve  as  a  vehicle  for  final  reports  of  work  performed  at 
NBS  under  the  sponsorship  of  other  government  agencies. 
Voluntary  Product  Standards — Developed  under  procedures 
published  by  the  Department  of  Commerce  in  Part  10, 
Title  15,  of  the  Code  of  Federal  Regulations.  The  purpose 
of  the  standards  is  to  establish  nationally  recognized  require- 
ments for  products,  and  to  provide  all  concerned  interests 
with  a  basis  for  common  understanding  of  the  characteristics 
of  the  products.  NBS  administers  this  program  as  a  supple- 
ment to  the  activities  of  the  private  sector  standardizing 
organizations. 

Consumer  Information  Series — Practical  information,  based 
on  NBS  research  and  experience,  covering  areas  of  interest 
to  the  consumer.  Easily  understandable  language  and 
illustrations  provide  useful  background  knowledge  for  shop- 
ping in  today's  technological  marketplace. 
Order  above  NBS  publications  from:  Superintendent  of 
Documents,  Government  Printing  Office,  Washington,  D.C. 
20402. 

Order  following  NBS  publications — NBSIR's  and  FIPS  from 
the  National  Technical  Information  Services,  Springfield, 
Va.  22161. 

Federal  Information  Processing  Standards  Publications 
(FIPS  PUB) — Publications  in  this  series  collectively  consti- 
tute the  Federal  Information  Processing  Standards  Register. 
Register  serves  as  the  official  source  of  information  in  the 
Federal  Government  regarding  standards  issued  by  NBS 
pursuant  to  the  Federal  Property  and  Administrative  Serv- 
ices Act  of  1949  as  amended.  Public  Law  89-306  (79  Stat. 
1127),  and  as  implemented  by  Executive  Order  11717 
(38  FR  12315,  dated  May  11,  1973)  and  Part  6  of  Title  15 
CFR  (Code  of  Federal  Regulations). 

NBS  Interagency  Reports  (NBSIR) — ^A  special  series  of 
interim  or  final  reports  on  work  performed  by  NBS  for 
outside  sponsors  (both  government  and  non-government). 
In  general,  initial  distribution  is  handled  by  the  sponsor; 
public  distribution  is  by  the  National  Technical  Information 
Services  (Springfield,  Va.  22161)  in  paper  copy  or  microfiche 
form. 


BIBLIOGRAPHIC  SUBSCRIPTION  SERVICES 


e  following  current-awareness  and  literature-survey  bibli- 
aphies  are  issued  periodically  by  the  Bureau: 
Cryogenic  Data  Center  Current  Awareness  Service.  A  litera- 
ture survey  issued  biweekly.  Annual  subscription:  Domes- 
tic, $25.00;  Foreign,  $30.00. 

iquified  Natural  Gas.  A  literature  survey  issued  quarterly, 
jnnual  subscription:  $20.00. 


Superconducting  Devices  and  Materials.  A  literature  survey 
issued  quarterly.  Annual  subscription:  $30.00.  Send  subscrip- 
tion orders  and  remittances  for  the  preceding  bibliographic 
services  to  National  Bureau  of  Standards,  Cryogenic  Data 
Center  (275.02)  Boulder,  Colorado  80302. 


U.S.  DEPARTMENT  OF  COMMERCE 
National  Bureau  of  Standards 

Washington.  D.C.  20234 


OFFICIAL  BUSINESS 

Penalty  for  Private  Use,  S300 


POSTAGE  AND  FEES  PAID 
U.S.  DEPARTMENT  OF  COMMERCE 
COM-219 


SPECIAL  FOURTH-CLASS  RATE 
BOOK 


i